app: expose server's default context length to UI

Parse the default_num_ctx from the server's "vram-based default context"
log line and expose it through the inference compute API. This eliminates
duplicate VRAM tier calculation logic in the frontend.

- Add InferenceInfo struct with Computes and DefaultContextLength
- Rename GetInferenceComputer to GetInferenceInfo
- Handle missing default context line gracefully (older servers)
- Add DefaultContextLength to InferenceComputeResponse
- Update Settings UI to use server's default, disable slider while loading
- Add disabled prop to Slider component (grays out + hides handle)
- Migrate existing users with context_length=4096 to 0 (auto mode)

README.md

@@ -358,6 +358,7 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [Odin Runes](https://github.com/leonid20000/OdinRunes)
 - [LLM-X](https://github.com/mrdjohnson/llm-x) (Progressive Web App)
 - [AnythingLLM (Docker + MacOs/Windows/Linux native app)](https://github.com/Mintplex-Labs/anything-llm)
+- [Screenpipe](https://github.com/mediar-ai/screenpipe) (24/7 screen & mic recording with AI-powered search, uses Ollama for local LLM features)
 - [Ollama Basic Chat: Uses HyperDiv Reactive UI](https://github.com/rapidarchitect/ollama_basic_chat)
 - [Ollama-chats RPG](https://github.com/drazdra/ollama-chats)
 - [IntelliBar](https://intellibar.app/) (AI-powered assistant for macOS)
@@ -465,6 +466,7 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [Clueless](https://github.com/KashyapTan/clueless) (Open Source & Local Cluely: A desktop application LLM assistant to help you talk to anything on your screen using locally served Ollama models. Also undetectable to screenshare)
 - [ollama-co2](https://github.com/carbonatedWaterOrg/ollama-co2) (FastAPI web interface for monitoring and managing local and remote Ollama servers with real-time model monitoring and concurrent downloads)
 - [Hillnote](https://hillnote.com) (A Markdown-first workspace designed to supercharge your AI workflow. Create documents ready to integrate with Claude, ChatGPT, Gemini, Cursor, and more - all while keeping your work on your device.)
+- [Stakpak](https://github.com/stakpak/agent) (An open source, vendor neutral DevOps agent that works with any model, and any stack, for teams who just want to ship)
 
 ### Cloud
 

app/server/server.go

@@ -41,6 +41,11 @@ type InferenceCompute struct {
 	VRAM    string
 }
 
+type InferenceInfo struct {
+	Computes             []InferenceCompute
+	DefaultContextLength int
+}
+
 func New(s *store.Store, devMode bool) *Server {
 	p := resolvePath("ollama")
 	return &Server{store: s, bin: p, dev: devMode}
@@ -262,9 +267,12 @@ func openRotatingLog() (io.WriteCloser, error) {
 
 // Attempt to retrieve inference compute information from the server
 // log.  Set ctx to timeout to control how long to wait for the logs to appear
-func GetInferenceComputer(ctx context.Context) ([]InferenceCompute, error) {
-	inference := []InferenceCompute{}
-	marker := regexp.MustCompile(`inference compute.*library=`)
+func GetInferenceInfo(ctx context.Context) (*InferenceInfo, error) {
+	info := &InferenceInfo{}
+	computeMarker := regexp.MustCompile(`inference compute.*library=`)
+	defaultCtxMarker := regexp.MustCompile(`vram-based default context`)
+	defaultCtxRegex := regexp.MustCompile(`default_num_ctx=(\d+)`)
+
 	q := `inference compute.*%s=["]([^"]*)["]`
 	nq := `inference compute.*%s=(\S+)\s`
 	type regex struct {
@@ -330,8 +338,8 @@ func GetInferenceComputer(ctx context.Context) ([]InferenceCompute, error) {
 		scanner := bufio.NewScanner(file)
 		for scanner.Scan() {
 			line := scanner.Text()
-			match := marker.FindStringSubmatch(line)
-			if len(match) > 0 {
+			// Check for inference compute lines
+			if computeMarker.MatchString(line) {
 				ic := InferenceCompute{
 					Library: get("library", line),
 					Variant: get("variant", line),
@@ -342,12 +350,25 @@ func GetInferenceComputer(ctx context.Context) ([]InferenceCompute, error) {
 				}
 
 				slog.Info("Matched", "inference compute", ic)
-				inference = append(inference, ic)
-			} else {
-				// Break out on first non matching line after we start matching
-				if len(inference) > 0 {
-					return inference, nil
+				info.Computes = append(info.Computes, ic)
+				continue
+			}
+			// Check for default context length line
+			if defaultCtxMarker.MatchString(line) {
+				match := defaultCtxRegex.FindStringSubmatch(line)
+				if len(match) > 1 {
+					numCtx, err := strconv.Atoi(match[1])
+					if err == nil {
+						info.DefaultContextLength = numCtx
+						slog.Info("Matched default context length", "default_num_ctx", numCtx)
+					}
 				}
+				return info, nil
+			}
+			// If we've found compute info but hit a non-matching line, return what we have
+			// This handles older server versions that don't log the default context line
+			if len(info.Computes) > 0 {
+				return info, nil
 			}
 		}
 		time.Sleep(100 * time.Millisecond)

app/server/server_test.go

@@ -136,44 +136,50 @@ func TestServerCmd(t *testing.T) {
 	}
 }
 
-func TestGetInferenceComputer(t *testing.T) {
+func TestGetInferenceInfo(t *testing.T) {
 	tests := []struct {
-		name string
-		log  string
-		exp  []InferenceCompute
+		name             string
+		log              string
+		expComputes      []InferenceCompute
+		expDefaultCtxLen int
 	}{
 		{
 			name: "metal",
 			log: `time=2025-06-30T09:23:07.374-07:00 level=DEBUG source=sched.go:108 msg="starting llm scheduler"
 time=2025-06-30T09:23:07.416-07:00 level=INFO source=types.go:130 msg="inference compute" id=0 library=metal variant="" compute="" driver=0.0 name="" total="96.0 GiB" available="96.0 GiB"
+time=2025-06-30T09:23:07.417-07:00 level=INFO source=routes.go:1721 msg="vram-based default context" total_vram="96.0 GiB" default_num_ctx=262144
 time=2025-06-30T09:25:56.197-07:00 level=DEBUG source=ggml.go:155 msg="key not found" key=general.alignment default=32
 `,
-			exp: []InferenceCompute{{
+			expComputes: []InferenceCompute{{
 				Library: "metal",
 				Driver:  "0.0",
 				VRAM:    "96.0 GiB",
 			}},
+			expDefaultCtxLen: 262144,
 		},
 		{
 			name: "cpu",
 			log: `time=2025-07-01T17:59:51.470Z level=INFO source=gpu.go:377 msg="no compatible GPUs were discovered"
 time=2025-07-01T17:59:51.470Z level=INFO source=types.go:130 msg="inference compute" id=0 library=cpu variant="" compute="" driver=0.0 name="" total="31.3 GiB" available="30.4 GiB"
+time=2025-07-01T17:59:51.471Z level=INFO source=routes.go:1721 msg="vram-based default context" total_vram="31.3 GiB" default_num_ctx=32768
 [GIN] 2025/07/01 - 18:00:09 | 200 |      50.263µs | 100.126.204.152 | HEAD     "/"
 `,
-			exp: []InferenceCompute{{
+			expComputes: []InferenceCompute{{
 				Library: "cpu",
 				Driver:  "0.0",
 				VRAM:    "31.3 GiB",
 			}},
+			expDefaultCtxLen: 32768,
 		},
 		{
 			name: "cuda1",
 			log: `time=2025-07-01T19:33:43.162Z level=DEBUG source=amd_linux.go:419 msg="amdgpu driver not detected /sys/module/amdgpu"
 releasing cuda driver library
 time=2025-07-01T19:33:43.162Z level=INFO source=types.go:130 msg="inference compute" id=GPU-452cac9f-6960-839c-4fb3-0cec83699196 library=cuda variant=v12 compute=6.1 driver=12.7 name="NVIDIA GeForce GT 1030" total="3.9 GiB" available="3.9 GiB"
+time=2025-07-01T19:33:43.163Z level=INFO source=routes.go:1721 msg="vram-based default context" total_vram="3.9 GiB" default_num_ctx=4096
 [GIN] 2025/07/01 - 18:00:09 | 200 |      50.263µs | 100.126.204.152 | HEAD     "/"
 `,
-			exp: []InferenceCompute{{
+			expComputes: []InferenceCompute{{
 				Library: "cuda",
 				Variant: "v12",
 				Compute: "6.1",
@@ -181,6 +187,7 @@ time=2025-07-01T19:33:43.162Z level=INFO source=types.go:130 msg="inference comp
 				Name:    "NVIDIA GeForce GT 1030",
 				VRAM:    "3.9 GiB",
 			}},
+			expDefaultCtxLen: 4096,
 		},
 		{
 			name: "frank",
@@ -188,9 +195,10 @@ time=2025-07-01T19:33:43.162Z level=INFO source=types.go:130 msg="inference comp
 		releasing cuda driver library
 		time=2025-07-01T19:36:13.315Z level=INFO source=types.go:130 msg="inference compute" id=GPU-d6de3398-9932-6902-11ec-fee8e424c8a2 library=cuda variant=v12 compute=7.5 driver=12.8 name="NVIDIA GeForce RTX 2080 Ti" total="10.6 GiB" available="10.4 GiB"
 		time=2025-07-01T19:36:13.315Z level=INFO source=types.go:130 msg="inference compute" id=GPU-9abb57639fa80c50 library=rocm variant="" compute=gfx1030 driver=6.3 name=1002:73bf total="16.0 GiB" available="1.3 GiB"
+		time=2025-07-01T19:36:13.316Z level=INFO source=routes.go:1721 msg="vram-based default context" total_vram="26.6 GiB" default_num_ctx=32768
 		[GIN] 2025/07/01 - 18:00:09 | 200 |      50.263µs | 100.126.204.152 | HEAD     "/"
 		`,
-			exp: []InferenceCompute{
+			expComputes: []InferenceCompute{
 				{
 					Library: "cuda",
 					Variant: "v12",
@@ -207,6 +215,20 @@ time=2025-07-01T19:33:43.162Z level=INFO source=types.go:130 msg="inference comp
 					VRAM:    "16.0 GiB",
 				},
 			},
+			expDefaultCtxLen: 32768,
+		},
+		{
+			name: "missing_default_context",
+			log: `time=2025-06-30T09:23:07.374-07:00 level=DEBUG source=sched.go:108 msg="starting llm scheduler"
+time=2025-06-30T09:23:07.416-07:00 level=INFO source=types.go:130 msg="inference compute" id=0 library=metal variant="" compute="" driver=0.0 name="" total="96.0 GiB" available="96.0 GiB"
+time=2025-06-30T09:25:56.197-07:00 level=DEBUG source=ggml.go:155 msg="key not found" key=general.alignment default=32
+`,
+			expComputes: []InferenceCompute{{
+				Library: "metal",
+				Driver:  "0.0",
+				VRAM:    "96.0 GiB",
+			}},
+			expDefaultCtxLen: 0, // No default context line, should return 0
 		},
 	}
 	for _, tt := range tests {
@@ -219,18 +241,21 @@ time=2025-07-01T19:33:43.162Z level=INFO source=types.go:130 msg="inference comp
 			}
 			ctx, cancel := context.WithTimeout(t.Context(), 10*time.Millisecond)
 			defer cancel()
-			ics, err := GetInferenceComputer(ctx)
+			info, err := GetInferenceInfo(ctx)
 			if err != nil {
-				t.Fatalf(" failed to get inference compute: %v", err)
+				t.Fatalf("failed to get inference info: %v", err)
 			}
-			if !reflect.DeepEqual(ics, tt.exp) {
-				t.Fatalf("got:\n%#v\nwant:\n%#v", ics, tt.exp)
+			if !reflect.DeepEqual(info.Computes, tt.expComputes) {
+				t.Fatalf("computes mismatch\ngot:\n%#v\nwant:\n%#v", info.Computes, tt.expComputes)
+			}
+			if info.DefaultContextLength != tt.expDefaultCtxLen {
+				t.Fatalf("default context length mismatch: got %d, want %d", info.DefaultContextLength, tt.expDefaultCtxLen)
 			}
 		})
 	}
 }
 
-func TestGetInferenceComputerTimeout(t *testing.T) {
+func TestGetInferenceInfoTimeout(t *testing.T) {
 	ctx, cancel := context.WithTimeout(t.Context(), 10*time.Millisecond)
 	defer cancel()
 	tmpDir := t.TempDir()
@@ -239,7 +264,7 @@ func TestGetInferenceComputerTimeout(t *testing.T) {
 	if err != nil {
 		t.Fatalf("failed to write log file %s: %s", serverLogPath, err)
 	}
-	_, err = GetInferenceComputer(ctx)
+	_, err = GetInferenceInfo(ctx)
 	if err == nil {
 		t.Fatal("expected timeout")
 	}

app/store/database.go

@@ -14,7 +14,7 @@ import (
 
 // currentSchemaVersion defines the current database schema version.
 // Increment this when making schema changes that require migrations.
-const currentSchemaVersion = 12
+const currentSchemaVersion = 13
 
 // database wraps the SQLite connection.
 // SQLite handles its own locking for concurrent access:
@@ -73,7 +73,7 @@ func (db *database) init() error {
 		agent BOOLEAN NOT NULL DEFAULT 0,
 		tools BOOLEAN NOT NULL DEFAULT 0,
 		working_dir TEXT NOT NULL DEFAULT '',
-		context_length INTEGER NOT NULL DEFAULT 4096,
+		context_length INTEGER NOT NULL DEFAULT 0,
 		window_width INTEGER NOT NULL DEFAULT 0,
 		window_height INTEGER NOT NULL DEFAULT 0,
 		config_migrated BOOLEAN NOT NULL DEFAULT 0,
@@ -244,6 +244,12 @@ func (db *database) migrate() error {
 				return fmt.Errorf("migrate v11 to v12: %w", err)
 			}
 			version = 12
+		case 12:
+			// change default context_length from 4096 to 0 (VRAM-based tiered defaults)
+			if err := db.migrateV12ToV13(); err != nil {
+				return fmt.Errorf("migrate v12 to v13: %w", err)
+			}
+			version = 13
 		default:
 			// If we have a version we don't recognize, just set it to current
 			// This might happen during development
@@ -452,6 +458,23 @@ func (db *database) migrateV11ToV12() error {
 	return nil
 }
 
+// migrateV12ToV13 changes the default context_length from 4096 to 0
+// When context_length is 0, the ollama server uses VRAM-based tiered defaults
+func (db *database) migrateV12ToV13() error {
+	// Update users who have the old default of 4096 to the new default of 0
+	_, err := db.conn.Exec(`UPDATE settings SET context_length = 0 WHERE context_length = 4096`)
+	if err != nil {
+		return fmt.Errorf("update context_length default: %w", err)
+	}
+
+	_, err = db.conn.Exec(`UPDATE settings SET schema_version = 13`)
+	if err != nil {
+		return fmt.Errorf("update schema version: %w", err)
+	}
+
+	return nil
+}
+
 // cleanupOrphanedData removes orphaned records that may exist due to the foreign key bug
 func (db *database) cleanupOrphanedData() error {
 	_, err := db.conn.Exec(`

app/store/testdata/schema.sql

@@ -13,7 +13,7 @@ CREATE TABLE IF NOT EXISTS settings (
     agent BOOLEAN NOT NULL DEFAULT 0,
     tools BOOLEAN NOT NULL DEFAULT 0,
     working_dir TEXT NOT NULL DEFAULT '',
-    context_length INTEGER NOT NULL DEFAULT 4096,
+    context_length INTEGER NOT NULL DEFAULT 0,
     window_width INTEGER NOT NULL DEFAULT 0,
     window_height INTEGER NOT NULL DEFAULT 0,
     config_migrated BOOLEAN NOT NULL DEFAULT 0,

app/ui/app/codegen/gotypes.gen.ts

@@ -289,10 +289,12 @@ export class InferenceCompute {
 }
 export class InferenceComputeResponse {
     inferenceComputes: InferenceCompute[];
+    defaultContextLength: number;
 
     constructor(source: any = {}) {
         if ('string' === typeof source) source = JSON.parse(source);
         this.inferenceComputes = this.convertValues(source["inferenceComputes"], InferenceCompute);
+        this.defaultContextLength = source["defaultContextLength"];
     }
 
 	convertValues(a: any, classs: any, asMap: boolean = false): any {

app/ui/app/src/api.ts

@@ -4,7 +4,6 @@ import {
   ChatEvent,
   DownloadEvent,
   ErrorEvent,
-  InferenceCompute,
   InferenceComputeResponse,
   ModelCapabilitiesResponse,
   Model,
@@ -379,7 +378,7 @@ export async function* pullModel(
   }
 }
 
-export async function getInferenceCompute(): Promise<InferenceCompute[]> {
+export async function getInferenceCompute(): Promise<InferenceComputeResponse> {
   const response = await fetch(`${API_BASE}/api/v1/inference-compute`);
   if (!response.ok) {
     throw new Error(
@@ -388,8 +387,7 @@ export async function getInferenceCompute(): Promise<InferenceCompute[]> {
   }
 
   const data = await response.json();
-  const inferenceComputeResponse = new InferenceComputeResponse(data);
-  return inferenceComputeResponse.inferenceComputes || [];
+  return new InferenceComputeResponse(data);
 }
 
 export async function fetchHealth(): Promise<boolean> {

app/ui/app/src/components/Settings.tsx

@@ -19,7 +19,7 @@ import { Settings as SettingsType } from "@/gotypes";
 import { useNavigate } from "@tanstack/react-router";
 import { useUser } from "@/hooks/useUser";
 import { useQuery, useMutation, useQueryClient } from "@tanstack/react-query";
-import { getSettings, updateSettings } from "@/api";
+import { getSettings, updateSettings, getInferenceCompute } from "@/api";
 
 function AnimatedDots() {
   return (
@@ -65,6 +65,13 @@ export default function Settings() {
 
   const settings = settingsData?.settings || null;
 
+  const { data: inferenceComputeResponse } = useQuery({
+    queryKey: ["inferenceCompute"],
+    queryFn: getInferenceCompute,
+  });
+
+  const defaultContextLength = inferenceComputeResponse?.defaultContextLength;
+
   const updateSettingsMutation = useMutation({
     mutationFn: updateSettings,
     onSuccess: () => {
@@ -148,7 +155,7 @@ export default function Settings() {
         Models: "",
         Agent: false,
         Tools: false,
-        ContextLength: 4096,
+        ContextLength: 0,
         AirplaneMode: false,
       });
       updateSettingsMutation.mutate(defaultSettings);
@@ -419,13 +426,11 @@ export default function Settings() {
                     </Description>
                     <div className="mt-3">
                       <Slider
-                        value={(() => {
-                          // Otherwise use the settings value
-                          return settings.ContextLength || 4096;
-                        })()}
+                        value={settings.ContextLength || defaultContextLength || 0}
                         onChange={(value) => {
                           handleChange("ContextLength", value);
                         }}
+                        disabled={!defaultContextLength}
                         options={[
                           { value: 4096, label: "4k" },
                           { value: 8192, label: "8k" },
@@ -440,6 +445,7 @@ export default function Settings() {
                   </div>
                 </div>
               </Field>
+
               {/* Airplane Mode */}
               <Field>
                 <div className="flex items-start justify-between gap-4">

app/ui/app/src/components/ui/slider.tsx

@@ -6,10 +6,11 @@ export interface SliderProps {
   value?: number;
   onChange?: (value: number) => void;
   className?: string;
+  disabled?: boolean;
 }
 
 const Slider = React.forwardRef<HTMLDivElement, SliderProps>(
-  ({ label, options, value = 0, onChange }, ref) => {
+  ({ label, options, value = 0, onChange, disabled = false }, ref) => {
     const [selectedValue, setSelectedValue] = React.useState(value);
     const [isDragging, setIsDragging] = React.useState(false);
     const containerRef = React.useRef<HTMLDivElement>(null);
@@ -20,6 +21,7 @@ const Slider = React.forwardRef<HTMLDivElement, SliderProps>(
     }, [value]);
 
     const handleClick = (optionValue: number) => {
+      if (disabled) return;
       setSelectedValue(optionValue);
       onChange?.(optionValue);
     };
@@ -39,6 +41,7 @@ const Slider = React.forwardRef<HTMLDivElement, SliderProps>(
     };
 
     const handleMouseDown = (e: React.MouseEvent) => {
+      if (disabled) return;
       setIsDragging(true);
       e.preventDefault();
     };
@@ -77,7 +80,7 @@ const Slider = React.forwardRef<HTMLDivElement, SliderProps>(
     }
 
     return (
-      <div className="space-y-2" ref={ref}>
+      <div className={`space-y-2 ${disabled ? "opacity-50" : ""}`} ref={ref}>
         {label && <label className="text-sm font-medium">{label}</label>}
         <div className="relative">
           <div className="absolute top-[9px] left-2 right-2 h-1 bg-neutral-200 dark:bg-neutral-700 pointer-events-none rounded-full" />
@@ -88,10 +91,11 @@ const Slider = React.forwardRef<HTMLDivElement, SliderProps>(
                 <button
                   onClick={() => handleClick(option.value)}
                   onMouseDown={handleMouseDown}
-                  className="relative px-3 py-6 -mx-3 -my-6 z-10 cursor-pointer"
+                  disabled={disabled}
+                  className={`relative px-3 py-6 -mx-3 -my-6 z-10 ${disabled ? "cursor-not-allowed" : "cursor-pointer"}`}
                 >
                   <div className="relative w-5 h-5 flex items-center justify-center">
-                    {selectedValue === option.value && (
+                    {selectedValue === option.value && !disabled && (
                       <div className="w-4 h-4 bg-white dark:bg-white border border-neutral-400 dark:border-neutral-500 rounded-full cursor-grab active:cursor-grabbing" />
                     )}
                   </div>

app/ui/app/src/hooks/useSelectedModel.ts

@@ -26,12 +26,14 @@ export function useSelectedModel(currentChatId?: string, searchQuery?: string) {
     currentChatId && currentChatId !== "new" ? currentChatId : "",
   );
 
-  const { data: inferenceComputes = [] } = useQuery({
-    queryKey: ["inference-compute"],
+  const { data: inferenceComputeResponse } = useQuery({
+    queryKey: ["inferenceCompute"],
     queryFn: getInferenceCompute,
     enabled: !settings.selectedModel, // Only fetch if no model is selected
   });
 
+  const inferenceComputes = inferenceComputeResponse?.inferenceComputes || [];
+
   const totalVRAM = useMemo(
     () => getTotalVRAM(inferenceComputes),
     [inferenceComputes],

app/ui/responses/types.go

@@ -45,7 +45,8 @@ type InferenceCompute struct {
 }
 
 type InferenceComputeResponse struct {
-	InferenceComputes []InferenceCompute `json:"inferenceComputes"`
+	InferenceComputes    []InferenceCompute `json:"inferenceComputes"`
+	DefaultContextLength int                `json:"defaultContextLength"`
 }
 
 type ModelCapabilitiesResponse struct {

app/ui/ui.go

@@ -1417,11 +1417,6 @@ func (s *Server) getSettings(w http.ResponseWriter, r *http.Request) error {
 		settings.Models = envconfig.Models()
 	}
 
-	// set default context length if not set
-	if settings.ContextLength == 0 {
-		settings.ContextLength = 4096
-	}
-
 	// Include current runtime settings
 	settings.Agent = s.Agent
 	settings.Tools = s.Tools
@@ -1463,14 +1458,14 @@ func (s *Server) settings(w http.ResponseWriter, r *http.Request) error {
 func (s *Server) getInferenceCompute(w http.ResponseWriter, r *http.Request) error {
 	ctx, cancel := context.WithTimeout(r.Context(), 500*time.Millisecond)
 	defer cancel()
-	serverInferenceComputes, err := server.GetInferenceComputer(ctx)
+	info, err := server.GetInferenceInfo(ctx)
 	if err != nil {
-		s.log().Error("failed to get inference compute", "error", err)
-		return fmt.Errorf("failed to get inference compute: %w", err)
+		s.log().Error("failed to get inference info", "error", err)
+		return fmt.Errorf("failed to get inference info: %w", err)
 	}
 
-	inferenceComputes := make([]responses.InferenceCompute, len(serverInferenceComputes))
-	for i, ic := range serverInferenceComputes {
+	inferenceComputes := make([]responses.InferenceCompute, len(info.Computes))
+	for i, ic := range info.Computes {
 		inferenceComputes[i] = responses.InferenceCompute{
 			Library: ic.Library,
 			Variant: ic.Variant,
@@ -1482,7 +1477,8 @@ func (s *Server) getInferenceCompute(w http.ResponseWriter, r *http.Request) err
 	}
 
 	response := responses.InferenceComputeResponse{
-		InferenceComputes: inferenceComputes,
+		InferenceComputes:    inferenceComputes,
+		DefaultContextLength: info.DefaultContextLength,
 	}
 
 	w.Header().Set("Content-Type", "application/json")

cmd/cmd.go

@@ -1419,10 +1419,10 @@ func thinkingOutputClosingText(plainText bool) string {
 	return readline.ColorGrey + readline.ColorBold + text + readline.ColorDefault
 }
 
-func chat(cmd *cobra.Command, opts runOptions) (*api.Message, *api.Metrics, error) {
+func chat(cmd *cobra.Command, opts runOptions) (*api.Message, error) {
 	client, err := api.ClientFromEnvironment()
 	if err != nil {
-		return nil, nil, err
+		return nil, err
 	}
 
 	p := progress.NewProgress(os.Stderr)
@@ -1515,7 +1515,7 @@ func chat(cmd *cobra.Command, opts runOptions) (*api.Message, *api.Metrics, erro
 
 	if err := client.Chat(cancelCtx, req, fn); err != nil {
 		if errors.Is(err, context.Canceled) {
-			return nil, nil, nil
+			return nil, nil
 		}
 
 		// this error should ideally be wrapped properly by the client
@@ -1523,9 +1523,9 @@ func chat(cmd *cobra.Command, opts runOptions) (*api.Message, *api.Metrics, erro
 			p.StopAndClear()
 			fmt.Println("An error occurred while processing your message. Please try again.")
 			fmt.Println()
-			return nil, nil, nil
+			return nil, nil
 		}
-		return nil, nil, err
+		return nil, err
 	}
 
 	if len(opts.Messages) > 0 {
@@ -1535,14 +1535,14 @@ func chat(cmd *cobra.Command, opts runOptions) (*api.Message, *api.Metrics, erro
 
 	verbose, err := cmd.Flags().GetBool("verbose")
 	if err != nil {
-		return nil, nil, err
+		return nil, err
 	}
 
 	if verbose {
 		latest.Summary()
 	}
 
-	return &api.Message{Role: role, Thinking: thinkingContent.String(), Content: fullResponse.String()}, &latest.Metrics, nil
+	return &api.Message{Role: role, Thinking: thinkingContent.String(), Content: fullResponse.String()}, nil
 }
 
 func generate(cmd *cobra.Command, opts runOptions) error {
@@ -1888,7 +1888,7 @@ func NewCLI() *cobra.Command {
 	serveCmd := &cobra.Command{
 		Use:     "serve",
 		Aliases: []string{"start"},
-		Short:   "Start ollama",
+		Short:   "Start Ollama",
 		Args:    cobra.ExactArgs(0),
 		RunE:    RunServer,
 	}

cmd/config/integrations.go

@@ -41,10 +41,18 @@ type Editor interface {
 // integrations is the registry of available integrations.
 var integrations = map[string]Runner{
 	"claude":   &Claude{},
-	"clawdbot": &Clawdbot{},
+	"clawdbot": &Openclaw{},
 	"codex":    &Codex{},
+	"moltbot":  &Openclaw{},
 	"droid":    &Droid{},
 	"opencode": &OpenCode{},
+	"openclaw": &Openclaw{},
+}
+
+// integrationAliases are hidden from the interactive selector but work as CLI arguments.
+var integrationAliases = map[string]bool{
+	"clawdbot": true,
+	"moltbot":  true,
 }
 
 func selectIntegration() (string, error) {
@@ -55,6 +63,9 @@ func selectIntegration() (string, error) {
 	names := slices.Sorted(maps.Keys(integrations))
 	var items []selectItem
 	for _, name := range names {
+		if integrationAliases[name] {
+			continue
+		}
 		r := integrations[name]
 		description := r.String()
 		if conn, err := loadIntegration(name); err == nil && len(conn.Models) > 0 {
@@ -243,10 +254,10 @@ func LaunchCmd(checkServerHeartbeat func(cmd *cobra.Command, args []string) erro
 
 Supported integrations:
   claude    Claude Code
-  clawdbot  Clawdbot
   codex     Codex
   droid     Droid
   opencode  OpenCode
+  openclaw  OpenClaw (aliases: clawdbot, moltbot)
 
 Examples:
   ollama launch

cmd/config/openclaw.go

@@ -13,26 +13,46 @@ import (
 	"github.com/ollama/ollama/envconfig"
 )
 
-type Clawdbot struct{}
+type Openclaw struct{}
 
-func (c *Clawdbot) String() string { return "Clawdbot" }
+func (c *Openclaw) String() string { return "OpenClaw" }
 
 const ansiGreen = "\033[32m"
 
-func (c *Clawdbot) Run(model string) error {
-	if _, err := exec.LookPath("clawdbot"); err != nil {
-		return fmt.Errorf("clawdbot is not installed, install from https://docs.clawd.bot")
+func (c *Openclaw) Run(model string) error {
+	bin := "openclaw"
+	if _, err := exec.LookPath(bin); err != nil {
+		bin = "clawdbot"
+		if _, err := exec.LookPath(bin); err != nil {
+			return fmt.Errorf("openclaw is not installed, install from https://docs.openclaw.ai")
+		}
 	}
 
 	models := []string{model}
-	if config, err := loadIntegration("clawdbot"); err == nil && len(config.Models) > 0 {
+	if config, err := loadIntegration("openclaw"); err == nil && len(config.Models) > 0 {
+		models = config.Models
+	} else if config, err := loadIntegration("clawdbot"); err == nil && len(config.Models) > 0 {
 		models = config.Models
 	}
 	if err := c.Edit(models); err != nil {
 		return fmt.Errorf("setup failed: %w", err)
 	}
 
-	cmd := exec.Command("clawdbot", "gateway")
+	if !c.onboarded() {
+		// Onboarding not completed: run it (model already set via Edit)
+		// Use "ollama" as gateway token for simple local access
+		cmd := exec.Command(bin, "onboard",
+			"--auth-choice", "skip",
+			"--gateway-token", "ollama",
+		)
+		cmd.Stdin = os.Stdin
+		cmd.Stdout = os.Stdout
+		cmd.Stderr = os.Stderr
+		return cmd.Run()
+	}
+
+	// Onboarding completed: run gateway
+	cmd := exec.Command(bin, "gateway")
 	cmd.Stdin = os.Stdin
 
 	// Capture output to detect "already running" message
@@ -42,22 +62,55 @@ func (c *Clawdbot) Run(model string) error {
 
 	err := cmd.Run()
 	if err != nil && strings.Contains(outputBuf.String(), "Gateway already running") {
-		fmt.Fprintf(os.Stderr, "%sClawdbot has been configured with Ollama. Gateway is already running.%s\n", ansiGreen, ansiReset)
+		fmt.Fprintf(os.Stderr, "%sOpenClaw has been configured with Ollama. Gateway is already running.%s\n", ansiGreen, ansiReset)
 		return nil
 	}
 	return err
 }
 
-func (c *Clawdbot) Paths() []string {
+// onboarded checks if OpenClaw onboarding wizard was completed
+// by looking for the wizard.lastRunAt marker in the config
+func (c *Openclaw) onboarded() bool {
+	home, err := os.UserHomeDir()
+	if err != nil {
+		return false
+	}
+
+	configPath := filepath.Join(home, ".openclaw", "openclaw.json")
+	legacyPath := filepath.Join(home, ".clawdbot", "clawdbot.json")
+
+	config := make(map[string]any)
+	if data, err := os.ReadFile(configPath); err == nil {
+		_ = json.Unmarshal(data, &config)
+	} else if data, err := os.ReadFile(legacyPath); err == nil {
+		_ = json.Unmarshal(data, &config)
+	} else {
+		return false
+	}
+
+	// Check for wizard.lastRunAt marker (set when onboarding completes)
+	wizard, _ := config["wizard"].(map[string]any)
+	if wizard == nil {
+		return false
+	}
+	lastRunAt, _ := wizard["lastRunAt"].(string)
+	return lastRunAt != ""
+}
+
+func (c *Openclaw) Paths() []string {
 	home, _ := os.UserHomeDir()
-	p := filepath.Join(home, ".clawdbot", "clawdbot.json")
+	p := filepath.Join(home, ".openclaw", "openclaw.json")
 	if _, err := os.Stat(p); err == nil {
 		return []string{p}
 	}
+	legacy := filepath.Join(home, ".clawdbot", "clawdbot.json")
+	if _, err := os.Stat(legacy); err == nil {
+		return []string{legacy}
+	}
 	return nil
 }
 
-func (c *Clawdbot) Edit(models []string) error {
+func (c *Openclaw) Edit(models []string) error {
 	if len(models) == 0 {
 		return nil
 	}
@@ -67,7 +120,8 @@ func (c *Clawdbot) Edit(models []string) error {
 		return err
 	}
 
-	configPath := filepath.Join(home, ".clawdbot", "clawdbot.json")
+	configPath := filepath.Join(home, ".openclaw", "openclaw.json")
+	legacyPath := filepath.Join(home, ".clawdbot", "clawdbot.json")
 	if err := os.MkdirAll(filepath.Dir(configPath), 0o755); err != nil {
 		return err
 	}
@@ -76,6 +130,8 @@ func (c *Clawdbot) Edit(models []string) error {
 	config := make(map[string]any)
 	if data, err := os.ReadFile(configPath); err == nil {
 		_ = json.Unmarshal(data, &config)
+	} else if data, err := os.ReadFile(legacyPath); err == nil {
+		_ = json.Unmarshal(data, &config)
 	}
 
 	// Navigate/create: models.providers.ollama (preserving other providers)
@@ -167,15 +223,18 @@ func (c *Clawdbot) Edit(models []string) error {
 	return writeWithBackup(configPath, data)
 }
 
-func (c *Clawdbot) Models() []string {
+func (c *Openclaw) Models() []string {
 	home, err := os.UserHomeDir()
 	if err != nil {
 		return nil
 	}
 
-	config, err := readJSONFile(filepath.Join(home, ".clawdbot", "clawdbot.json"))
+	config, err := readJSONFile(filepath.Join(home, ".openclaw", "openclaw.json"))
 	if err != nil {
-		return nil
+		config, err = readJSONFile(filepath.Join(home, ".clawdbot", "clawdbot.json"))
+		if err != nil {
+			return nil
+		}
 	}
 
 	modelsSection, _ := config["models"].(map[string]any)

cmd/config/openclaw_test.go

@@ -8,12 +8,12 @@ import (
 	"testing"
 )
 
-func TestClawdbotIntegration(t *testing.T) {
-	c := &Clawdbot{}
+func TestOpenclawIntegration(t *testing.T) {
+	c := &Openclaw{}
 
 	t.Run("String", func(t *testing.T) {
-		if got := c.String(); got != "Clawdbot" {
-			t.Errorf("String() = %q, want %q", got, "Clawdbot")
+		if got := c.String(); got != "OpenClaw" {
+			t.Errorf("String() = %q, want %q", got, "OpenClaw")
 		}
 	})
 
@@ -26,13 +26,13 @@ func TestClawdbotIntegration(t *testing.T) {
 	})
 }
 
-func TestClawdbotEdit(t *testing.T) {
-	c := &Clawdbot{}
+func TestOpenclawEdit(t *testing.T) {
+	c := &Openclaw{}
 	tmpDir := t.TempDir()
 	setTestHome(t, tmpDir)
 
-	configDir := filepath.Join(tmpDir, ".clawdbot")
-	configPath := filepath.Join(configDir, "clawdbot.json")
+	configDir := filepath.Join(tmpDir, ".openclaw")
+	configPath := filepath.Join(configDir, "openclaw.json")
 
 	cleanup := func() { os.RemoveAll(configDir) }
 
@@ -41,8 +41,8 @@ func TestClawdbotEdit(t *testing.T) {
 		if err := c.Edit([]string{"llama3.2"}); err != nil {
 			t.Fatal(err)
 		}
-		assertClawdbotModelExists(t, configPath, "llama3.2")
-		assertClawdbotPrimaryModel(t, configPath, "ollama/llama3.2")
+		assertOpenclawModelExists(t, configPath, "llama3.2")
+		assertOpenclawPrimaryModel(t, configPath, "ollama/llama3.2")
 	})
 
 	t.Run("multiple models - first is primary", func(t *testing.T) {
@@ -50,9 +50,9 @@ func TestClawdbotEdit(t *testing.T) {
 		if err := c.Edit([]string{"llama3.2", "mistral"}); err != nil {
 			t.Fatal(err)
 		}
-		assertClawdbotModelExists(t, configPath, "llama3.2")
-		assertClawdbotModelExists(t, configPath, "mistral")
-		assertClawdbotPrimaryModel(t, configPath, "ollama/llama3.2")
+		assertOpenclawModelExists(t, configPath, "llama3.2")
+		assertOpenclawModelExists(t, configPath, "mistral")
+		assertOpenclawPrimaryModel(t, configPath, "ollama/llama3.2")
 	})
 
 	t.Run("preserve other providers", func(t *testing.T) {
@@ -127,8 +127,8 @@ func TestClawdbotEdit(t *testing.T) {
 		c.Edit([]string{"llama3.2", "mistral"})
 		c.Edit([]string{"llama3.2"})
 
-		assertClawdbotModelExists(t, configPath, "llama3.2")
-		assertClawdbotModelNotExists(t, configPath, "mistral")
+		assertOpenclawModelExists(t, configPath, "llama3.2")
+		assertOpenclawModelNotExists(t, configPath, "mistral")
 	})
 
 	t.Run("empty models is no-op", func(t *testing.T) {
@@ -169,12 +169,12 @@ func TestClawdbotEdit(t *testing.T) {
 		if err := c.Edit([]string{"llama3.2"}); err != nil {
 			t.Fatal(err)
 		}
-		assertClawdbotModelExists(t, configPath, "llama3.2")
+		assertOpenclawModelExists(t, configPath, "llama3.2")
 	})
 }
 
-func TestClawdbotModels(t *testing.T) {
-	c := &Clawdbot{}
+func TestOpenclawModels(t *testing.T) {
+	c := &Openclaw{}
 	tmpDir := t.TempDir()
 	setTestHome(t, tmpDir)
 
@@ -185,9 +185,9 @@ func TestClawdbotModels(t *testing.T) {
 	})
 
 	t.Run("returns all ollama models", func(t *testing.T) {
-		configDir := filepath.Join(tmpDir, ".clawdbot")
+		configDir := filepath.Join(tmpDir, ".openclaw")
 		os.MkdirAll(configDir, 0o755)
-		os.WriteFile(filepath.Join(configDir, "clawdbot.json"), []byte(`{
+		os.WriteFile(filepath.Join(configDir, "openclaw.json"), []byte(`{
 			"models":{"providers":{"ollama":{"models":[
 				{"id":"llama3.2"},
 				{"id":"mistral"}
@@ -202,7 +202,7 @@ func TestClawdbotModels(t *testing.T) {
 }
 
 // Helper functions
-func assertClawdbotModelExists(t *testing.T, path, model string) {
+func assertOpenclawModelExists(t *testing.T, path, model string) {
 	t.Helper()
 	data, _ := os.ReadFile(path)
 	var cfg map[string]any
@@ -221,7 +221,7 @@ func assertClawdbotModelExists(t *testing.T, path, model string) {
 	t.Errorf("model %s not found", model)
 }
 
-func assertClawdbotModelNotExists(t *testing.T, path, model string) {
+func assertOpenclawModelNotExists(t *testing.T, path, model string) {
 	t.Helper()
 	data, _ := os.ReadFile(path)
 	var cfg map[string]any
@@ -239,7 +239,7 @@ func assertClawdbotModelNotExists(t *testing.T, path, model string) {
 	}
 }
 
-func assertClawdbotPrimaryModel(t *testing.T, path, expected string) {
+func assertOpenclawPrimaryModel(t *testing.T, path, expected string) {
 	t.Helper()
 	data, _ := os.ReadFile(path)
 	var cfg map[string]any
@@ -252,15 +252,15 @@ func assertClawdbotPrimaryModel(t *testing.T, path, expected string) {
 	}
 }
 
-func TestClawdbotPaths(t *testing.T) {
-	c := &Clawdbot{}
+func TestOpenclawPaths(t *testing.T) {
+	c := &Openclaw{}
 
 	t.Run("returns path when config exists", func(t *testing.T) {
 		tmpDir := t.TempDir()
 		setTestHome(t, tmpDir)
-		configDir := filepath.Join(tmpDir, ".clawdbot")
+		configDir := filepath.Join(tmpDir, ".openclaw")
 		os.MkdirAll(configDir, 0o755)
-		os.WriteFile(filepath.Join(configDir, "clawdbot.json"), []byte(`{}`), 0o644)
+		os.WriteFile(filepath.Join(configDir, "openclaw.json"), []byte(`{}`), 0o644)
 
 		paths := c.Paths()
 		if len(paths) != 1 {
@@ -277,12 +277,12 @@ func TestClawdbotPaths(t *testing.T) {
 	})
 }
 
-func TestClawdbotModelsEdgeCases(t *testing.T) {
-	c := &Clawdbot{}
+func TestOpenclawModelsEdgeCases(t *testing.T) {
+	c := &Openclaw{}
 	tmpDir := t.TempDir()
 	setTestHome(t, tmpDir)
-	configDir := filepath.Join(tmpDir, ".clawdbot")
-	configPath := filepath.Join(configDir, "clawdbot.json")
+	configDir := filepath.Join(tmpDir, ".openclaw")
+	configPath := filepath.Join(configDir, "openclaw.json")
 	cleanup := func() { os.RemoveAll(configDir) }
 
 	t.Run("corrupted JSON returns nil", func(t *testing.T) {
@@ -340,11 +340,11 @@ func TestClawdbotModelsEdgeCases(t *testing.T) {
 	})
 }
 
-func TestClawdbotEditSchemaFields(t *testing.T) {
-	c := &Clawdbot{}
+func TestOpenclawEditSchemaFields(t *testing.T) {
+	c := &Openclaw{}
 	tmpDir := t.TempDir()
 	setTestHome(t, tmpDir)
-	configPath := filepath.Join(tmpDir, ".clawdbot", "clawdbot.json")
+	configPath := filepath.Join(tmpDir, ".openclaw", "openclaw.json")
 
 	if err := c.Edit([]string{"llama3.2"}); err != nil {
 		t.Fatal(err)
@@ -381,20 +381,20 @@ func TestClawdbotEditSchemaFields(t *testing.T) {
 	}
 }
 
-func TestClawdbotEditModelNames(t *testing.T) {
-	c := &Clawdbot{}
+func TestOpenclawEditModelNames(t *testing.T) {
+	c := &Openclaw{}
 	tmpDir := t.TempDir()
 	setTestHome(t, tmpDir)
-	configPath := filepath.Join(tmpDir, ".clawdbot", "clawdbot.json")
-	cleanup := func() { os.RemoveAll(filepath.Join(tmpDir, ".clawdbot")) }
+	configPath := filepath.Join(tmpDir, ".openclaw", "openclaw.json")
+	cleanup := func() { os.RemoveAll(filepath.Join(tmpDir, ".openclaw")) }
 
 	t.Run("model with colon tag", func(t *testing.T) {
 		cleanup()
 		if err := c.Edit([]string{"llama3.2:70b"}); err != nil {
 			t.Fatal(err)
 		}
-		assertClawdbotModelExists(t, configPath, "llama3.2:70b")
-		assertClawdbotPrimaryModel(t, configPath, "ollama/llama3.2:70b")
+		assertOpenclawModelExists(t, configPath, "llama3.2:70b")
+		assertOpenclawPrimaryModel(t, configPath, "ollama/llama3.2:70b")
 	})
 
 	t.Run("model with slash", func(t *testing.T) {
@@ -402,8 +402,8 @@ func TestClawdbotEditModelNames(t *testing.T) {
 		if err := c.Edit([]string{"library/model:tag"}); err != nil {
 			t.Fatal(err)
 		}
-		assertClawdbotModelExists(t, configPath, "library/model:tag")
-		assertClawdbotPrimaryModel(t, configPath, "ollama/library/model:tag")
+		assertOpenclawModelExists(t, configPath, "library/model:tag")
+		assertOpenclawPrimaryModel(t, configPath, "ollama/library/model:tag")
 	})
 
 	t.Run("model with hyphen", func(t *testing.T) {
@@ -411,16 +411,16 @@ func TestClawdbotEditModelNames(t *testing.T) {
 		if err := c.Edit([]string{"test-model"}); err != nil {
 			t.Fatal(err)
 		}
-		assertClawdbotModelExists(t, configPath, "test-model")
+		assertOpenclawModelExists(t, configPath, "test-model")
 	})
 }
 
-func TestClawdbotEditAgentsPreservation(t *testing.T) {
-	c := &Clawdbot{}
+func TestOpenclawEditAgentsPreservation(t *testing.T) {
+	c := &Openclaw{}
 	tmpDir := t.TempDir()
 	setTestHome(t, tmpDir)
-	configDir := filepath.Join(tmpDir, ".clawdbot")
-	configPath := filepath.Join(configDir, "clawdbot.json")
+	configDir := filepath.Join(tmpDir, ".openclaw")
+	configPath := filepath.Join(configDir, "openclaw.json")
 	cleanup := func() { os.RemoveAll(configDir) }
 
 	t.Run("preserve other agent defaults", func(t *testing.T) {
@@ -457,7 +457,7 @@ func TestClawdbotEditAgentsPreservation(t *testing.T) {
 	})
 }
 
-const testClawdbotFixture = `{
+const testOpenclawFixture = `{
   "theme": "dark",
   "mcp": {"servers": {"custom": {"enabled": true}}},
   "models": {
@@ -475,15 +475,15 @@ const testClawdbotFixture = `{
   }
 }`
 
-func TestClawdbotEdit_RoundTrip(t *testing.T) {
-	c := &Clawdbot{}
+func TestOpenclawEdit_RoundTrip(t *testing.T) {
+	c := &Openclaw{}
 	tmpDir := t.TempDir()
 	setTestHome(t, tmpDir)
-	configDir := filepath.Join(tmpDir, ".clawdbot")
-	configPath := filepath.Join(configDir, "clawdbot.json")
+	configDir := filepath.Join(tmpDir, ".openclaw")
+	configPath := filepath.Join(configDir, "openclaw.json")
 
 	os.MkdirAll(configDir, 0o755)
-	os.WriteFile(configPath, []byte(testClawdbotFixture), 0o644)
+	os.WriteFile(configPath, []byte(testOpenclawFixture), 0o644)
 
 	if err := c.Edit([]string{"llama3.2", "mistral"}); err != nil {
 		t.Fatal(err)
@@ -521,15 +521,15 @@ func TestClawdbotEdit_RoundTrip(t *testing.T) {
 	}
 }
 
-func TestClawdbotEdit_Idempotent(t *testing.T) {
-	c := &Clawdbot{}
+func TestOpenclawEdit_Idempotent(t *testing.T) {
+	c := &Openclaw{}
 	tmpDir := t.TempDir()
 	setTestHome(t, tmpDir)
-	configDir := filepath.Join(tmpDir, ".clawdbot")
-	configPath := filepath.Join(configDir, "clawdbot.json")
+	configDir := filepath.Join(tmpDir, ".openclaw")
+	configPath := filepath.Join(configDir, "openclaw.json")
 
 	os.MkdirAll(configDir, 0o755)
-	os.WriteFile(configPath, []byte(testClawdbotFixture), 0o644)
+	os.WriteFile(configPath, []byte(testOpenclawFixture), 0o644)
 
 	c.Edit([]string{"llama3.2", "mistral"})
 	firstData, _ := os.ReadFile(configPath)
@@ -542,15 +542,15 @@ func TestClawdbotEdit_Idempotent(t *testing.T) {
 	}
 }
 
-func TestClawdbotEdit_MultipleConsecutiveEdits(t *testing.T) {
-	c := &Clawdbot{}
+func TestOpenclawEdit_MultipleConsecutiveEdits(t *testing.T) {
+	c := &Openclaw{}
 	tmpDir := t.TempDir()
 	setTestHome(t, tmpDir)
-	configDir := filepath.Join(tmpDir, ".clawdbot")
-	configPath := filepath.Join(configDir, "clawdbot.json")
+	configDir := filepath.Join(tmpDir, ".openclaw")
+	configPath := filepath.Join(configDir, "openclaw.json")
 
 	os.MkdirAll(configDir, 0o755)
-	os.WriteFile(configPath, []byte(testClawdbotFixture), 0o644)
+	os.WriteFile(configPath, []byte(testOpenclawFixture), 0o644)
 
 	for i := range 10 {
 		models := []string{"model-a", "model-b"}
@@ -573,12 +573,12 @@ func TestClawdbotEdit_MultipleConsecutiveEdits(t *testing.T) {
 	}
 }
 
-func TestClawdbotEdit_BackupCreated(t *testing.T) {
-	c := &Clawdbot{}
+func TestOpenclawEdit_BackupCreated(t *testing.T) {
+	c := &Openclaw{}
 	tmpDir := t.TempDir()
 	setTestHome(t, tmpDir)
-	configDir := filepath.Join(tmpDir, ".clawdbot")
-	configPath := filepath.Join(configDir, "clawdbot.json")
+	configDir := filepath.Join(tmpDir, ".openclaw")
+	configPath := filepath.Join(configDir, "openclaw.json")
 	backupDir := filepath.Join(os.TempDir(), "ollama-backups")
 
 	os.MkdirAll(configDir, 0o755)
@@ -590,7 +590,7 @@ func TestClawdbotEdit_BackupCreated(t *testing.T) {
 		t.Fatal(err)
 	}
 
-	backups, _ := filepath.Glob(filepath.Join(backupDir, "clawdbot.json.*"))
+	backups, _ := filepath.Glob(filepath.Join(backupDir, "openclaw.json.*"))
 	foundBackup := false
 	for _, backup := range backups {
 		data, _ := os.ReadFile(backup)
@@ -605,11 +605,151 @@ func TestClawdbotEdit_BackupCreated(t *testing.T) {
 	}
 }
 
-func TestClawdbotEdit_CreatesDirectoryIfMissing(t *testing.T) {
-	c := &Clawdbot{}
+func TestOpenclawClawdbotAlias(t *testing.T) {
+	for _, alias := range []string{"clawdbot", "moltbot"} {
+		t.Run(alias+" alias resolves to Openclaw runner", func(t *testing.T) {
+			r, ok := integrations[alias]
+			if !ok {
+				t.Fatalf("%s not found in integrations", alias)
+			}
+			if _, ok := r.(*Openclaw); !ok {
+				t.Errorf("%s integration is %T, want *Openclaw", alias, r)
+			}
+		})
+
+		t.Run(alias+" is hidden from selector", func(t *testing.T) {
+			if !integrationAliases[alias] {
+				t.Errorf("%s should be in integrationAliases", alias)
+			}
+		})
+	}
+}
+
+func TestOpenclawLegacyPaths(t *testing.T) {
+	c := &Openclaw{}
+
+	t.Run("falls back to legacy clawdbot path", func(t *testing.T) {
+		tmpDir := t.TempDir()
+		setTestHome(t, tmpDir)
+		legacyDir := filepath.Join(tmpDir, ".clawdbot")
+		os.MkdirAll(legacyDir, 0o755)
+		os.WriteFile(filepath.Join(legacyDir, "clawdbot.json"), []byte(`{}`), 0o644)
+
+		paths := c.Paths()
+		if len(paths) != 1 {
+			t.Fatalf("expected 1 path, got %d", len(paths))
+		}
+		if paths[0] != filepath.Join(legacyDir, "clawdbot.json") {
+			t.Errorf("expected legacy path, got %s", paths[0])
+		}
+	})
+
+	t.Run("prefers new path over legacy", func(t *testing.T) {
+		tmpDir := t.TempDir()
+		setTestHome(t, tmpDir)
+		newDir := filepath.Join(tmpDir, ".openclaw")
+		legacyDir := filepath.Join(tmpDir, ".clawdbot")
+		os.MkdirAll(newDir, 0o755)
+		os.MkdirAll(legacyDir, 0o755)
+		os.WriteFile(filepath.Join(newDir, "openclaw.json"), []byte(`{}`), 0o644)
+		os.WriteFile(filepath.Join(legacyDir, "clawdbot.json"), []byte(`{}`), 0o644)
+
+		paths := c.Paths()
+		if len(paths) != 1 {
+			t.Fatalf("expected 1 path, got %d", len(paths))
+		}
+		if paths[0] != filepath.Join(newDir, "openclaw.json") {
+			t.Errorf("expected new path, got %s", paths[0])
+		}
+	})
+
+	t.Run("Models reads from legacy path", func(t *testing.T) {
+		tmpDir := t.TempDir()
+		setTestHome(t, tmpDir)
+		legacyDir := filepath.Join(tmpDir, ".clawdbot")
+		os.MkdirAll(legacyDir, 0o755)
+		os.WriteFile(filepath.Join(legacyDir, "clawdbot.json"), []byte(`{
+			"models":{"providers":{"ollama":{"models":[{"id":"llama3.2"}]}}}
+		}`), 0o644)
+
+		models := c.Models()
+		if len(models) != 1 || models[0] != "llama3.2" {
+			t.Errorf("expected [llama3.2], got %v", models)
+		}
+	})
+
+	t.Run("Models prefers new path over legacy", func(t *testing.T) {
+		tmpDir := t.TempDir()
+		setTestHome(t, tmpDir)
+		newDir := filepath.Join(tmpDir, ".openclaw")
+		legacyDir := filepath.Join(tmpDir, ".clawdbot")
+		os.MkdirAll(newDir, 0o755)
+		os.MkdirAll(legacyDir, 0o755)
+		os.WriteFile(filepath.Join(newDir, "openclaw.json"), []byte(`{
+			"models":{"providers":{"ollama":{"models":[{"id":"new-model"}]}}}
+		}`), 0o644)
+		os.WriteFile(filepath.Join(legacyDir, "clawdbot.json"), []byte(`{
+			"models":{"providers":{"ollama":{"models":[{"id":"legacy-model"}]}}}
+		}`), 0o644)
+
+		models := c.Models()
+		if len(models) != 1 || models[0] != "new-model" {
+			t.Errorf("expected [new-model], got %v", models)
+		}
+	})
+
+	t.Run("Edit reads new path over legacy when both exist", func(t *testing.T) {
+		tmpDir := t.TempDir()
+		setTestHome(t, tmpDir)
+		newDir := filepath.Join(tmpDir, ".openclaw")
+		legacyDir := filepath.Join(tmpDir, ".clawdbot")
+		os.MkdirAll(newDir, 0o755)
+		os.MkdirAll(legacyDir, 0o755)
+		os.WriteFile(filepath.Join(newDir, "openclaw.json"), []byte(`{"theme":"new"}`), 0o644)
+		os.WriteFile(filepath.Join(legacyDir, "clawdbot.json"), []byte(`{"theme":"legacy"}`), 0o644)
+
+		if err := c.Edit([]string{"llama3.2"}); err != nil {
+			t.Fatal(err)
+		}
+
+		data, _ := os.ReadFile(filepath.Join(newDir, "openclaw.json"))
+		var cfg map[string]any
+		json.Unmarshal(data, &cfg)
+		if cfg["theme"] != "new" {
+			t.Errorf("expected theme from new config, got %v", cfg["theme"])
+		}
+	})
+
+	t.Run("Edit migrates from legacy config", func(t *testing.T) {
+		tmpDir := t.TempDir()
+		setTestHome(t, tmpDir)
+		legacyDir := filepath.Join(tmpDir, ".clawdbot")
+		os.MkdirAll(legacyDir, 0o755)
+		os.WriteFile(filepath.Join(legacyDir, "clawdbot.json"), []byte(`{"theme":"dark"}`), 0o644)
+
+		if err := c.Edit([]string{"llama3.2"}); err != nil {
+			t.Fatal(err)
+		}
+
+		// Should write to new path
+		newPath := filepath.Join(tmpDir, ".openclaw", "openclaw.json")
+		data, err := os.ReadFile(newPath)
+		if err != nil {
+			t.Fatal("expected new config file to be created")
+		}
+		var cfg map[string]any
+		json.Unmarshal(data, &cfg)
+		if cfg["theme"] != "dark" {
+			t.Error("legacy theme setting was not migrated")
+		}
+	})
+}
+
+func TestOpenclawEdit_CreatesDirectoryIfMissing(t *testing.T) {
+	c := &Openclaw{}
 	tmpDir := t.TempDir()
 	setTestHome(t, tmpDir)
-	configDir := filepath.Join(tmpDir, ".clawdbot")
+	configDir := filepath.Join(tmpDir, ".openclaw")
 
 	if _, err := os.Stat(configDir); !os.IsNotExist(err) {
 		t.Fatal("directory should not exist before test")
@@ -623,3 +763,116 @@ func TestClawdbotEdit_CreatesDirectoryIfMissing(t *testing.T) {
 		t.Fatal("directory was not created")
 	}
 }
+
+func TestOpenclawOnboarded(t *testing.T) {
+	c := &Openclaw{}
+
+	t.Run("returns false when no config exists", func(t *testing.T) {
+		tmpDir := t.TempDir()
+		setTestHome(t, tmpDir)
+		if c.onboarded() {
+			t.Error("expected false when no config exists")
+		}
+	})
+
+	t.Run("returns false when config exists but no wizard section", func(t *testing.T) {
+		tmpDir := t.TempDir()
+		setTestHome(t, tmpDir)
+		configDir := filepath.Join(tmpDir, ".openclaw")
+		os.MkdirAll(configDir, 0o755)
+		os.WriteFile(filepath.Join(configDir, "openclaw.json"), []byte(`{"theme":"dark"}`), 0o644)
+
+		if c.onboarded() {
+			t.Error("expected false when no wizard section")
+		}
+	})
+
+	t.Run("returns false when wizard section exists but no lastRunAt", func(t *testing.T) {
+		tmpDir := t.TempDir()
+		setTestHome(t, tmpDir)
+		configDir := filepath.Join(tmpDir, ".openclaw")
+		os.MkdirAll(configDir, 0o755)
+		os.WriteFile(filepath.Join(configDir, "openclaw.json"), []byte(`{"wizard":{}}`), 0o644)
+
+		if c.onboarded() {
+			t.Error("expected false when wizard.lastRunAt is missing")
+		}
+	})
+
+	t.Run("returns false when wizard.lastRunAt is empty string", func(t *testing.T) {
+		tmpDir := t.TempDir()
+		setTestHome(t, tmpDir)
+		configDir := filepath.Join(tmpDir, ".openclaw")
+		os.MkdirAll(configDir, 0o755)
+		os.WriteFile(filepath.Join(configDir, "openclaw.json"), []byte(`{"wizard":{"lastRunAt":""}}`), 0o644)
+
+		if c.onboarded() {
+			t.Error("expected false when wizard.lastRunAt is empty")
+		}
+	})
+
+	t.Run("returns true when wizard.lastRunAt is set", func(t *testing.T) {
+		tmpDir := t.TempDir()
+		setTestHome(t, tmpDir)
+		configDir := filepath.Join(tmpDir, ".openclaw")
+		os.MkdirAll(configDir, 0o755)
+		os.WriteFile(filepath.Join(configDir, "openclaw.json"), []byte(`{"wizard":{"lastRunAt":"2024-01-01T00:00:00Z"}}`), 0o644)
+
+		if !c.onboarded() {
+			t.Error("expected true when wizard.lastRunAt is set")
+		}
+	})
+
+	t.Run("checks legacy clawdbot path", func(t *testing.T) {
+		tmpDir := t.TempDir()
+		setTestHome(t, tmpDir)
+		legacyDir := filepath.Join(tmpDir, ".clawdbot")
+		os.MkdirAll(legacyDir, 0o755)
+		os.WriteFile(filepath.Join(legacyDir, "clawdbot.json"), []byte(`{"wizard":{"lastRunAt":"2024-01-01T00:00:00Z"}}`), 0o644)
+
+		if !c.onboarded() {
+			t.Error("expected true when legacy config has wizard.lastRunAt")
+		}
+	})
+
+	t.Run("prefers new path over legacy", func(t *testing.T) {
+		tmpDir := t.TempDir()
+		setTestHome(t, tmpDir)
+		newDir := filepath.Join(tmpDir, ".openclaw")
+		legacyDir := filepath.Join(tmpDir, ".clawdbot")
+		os.MkdirAll(newDir, 0o755)
+		os.MkdirAll(legacyDir, 0o755)
+		// New path has no wizard marker
+		os.WriteFile(filepath.Join(newDir, "openclaw.json"), []byte(`{}`), 0o644)
+		// Legacy has wizard marker
+		os.WriteFile(filepath.Join(legacyDir, "clawdbot.json"), []byte(`{"wizard":{"lastRunAt":"2024-01-01T00:00:00Z"}}`), 0o644)
+
+		if c.onboarded() {
+			t.Error("expected false - should prefer new path which has no wizard marker")
+		}
+	})
+
+	t.Run("handles corrupted JSON gracefully", func(t *testing.T) {
+		tmpDir := t.TempDir()
+		setTestHome(t, tmpDir)
+		configDir := filepath.Join(tmpDir, ".openclaw")
+		os.MkdirAll(configDir, 0o755)
+		os.WriteFile(filepath.Join(configDir, "openclaw.json"), []byte(`{corrupted`), 0o644)
+
+		if c.onboarded() {
+			t.Error("expected false for corrupted JSON")
+		}
+	})
+
+	t.Run("handles wrong type for wizard section", func(t *testing.T) {
+		tmpDir := t.TempDir()
+		setTestHome(t, tmpDir)
+		configDir := filepath.Join(tmpDir, ".openclaw")
+		os.MkdirAll(configDir, 0o755)
+		os.WriteFile(filepath.Join(configDir, "openclaw.json"), []byte(`{"wizard":"not a map"}`), 0o644)
+
+		if c.onboarded() {
+			t.Error("expected false when wizard is wrong type")
+		}
+	})
+}

cmd/config/selector.go

@@ -275,7 +275,11 @@ func parseInput(r io.Reader) (inputEvent, byte, error) {
 func renderSelect(w io.Writer, prompt string, s *selectState) int {
 	filtered := s.filtered()
 
-	fmt.Fprintf(w, "%s %s\r\n", prompt, s.filter)
+	if s.filter == "" {
+		fmt.Fprintf(w, "%s %sType to filter...%s\r\n", prompt, ansiGray, ansiReset)
+	} else {
+		fmt.Fprintf(w, "%s %s\r\n", prompt, s.filter)
+	}
 	lineCount := 1
 
 	if len(filtered) == 0 {
@@ -314,7 +318,11 @@ func renderSelect(w io.Writer, prompt string, s *selectState) int {
 func renderMultiSelect(w io.Writer, prompt string, s *multiSelectState) int {
 	filtered := s.filtered()
 
-	fmt.Fprintf(w, "%s %s\r\n", prompt, s.filter)
+	if s.filter == "" {
+		fmt.Fprintf(w, "%s %sType to filter...%s\r\n", prompt, ansiGray, ansiReset)
+	} else {
+		fmt.Fprintf(w, "%s %s\r\n", prompt, s.filter)
+	}
 	lineCount := 1
 
 	if len(filtered) == 0 {

cmd/interactive.go

@@ -30,9 +30,6 @@ const (
 )
 
 func generateInteractive(cmd *cobra.Command, opts runOptions) error {
-	var sessionPromptTokens int64
-	var sessionCompletionTokens int64
-
 	usage := func() {
 		fmt.Fprintln(os.Stderr, "Available Commands:")
 		fmt.Fprintln(os.Stderr, "  /set            Set session variables")
@@ -40,7 +37,6 @@ func generateInteractive(cmd *cobra.Command, opts runOptions) error {
 		fmt.Fprintln(os.Stderr, "  /load <model>   Load a session or model")
 		fmt.Fprintln(os.Stderr, "  /save <model>   Save your current session")
 		fmt.Fprintln(os.Stderr, "  /clear          Clear session context")
-		fmt.Fprintln(os.Stderr, "  /usage          Show session token usage")
 		fmt.Fprintln(os.Stderr, "  /bye            Exit")
 		fmt.Fprintln(os.Stderr, "  /?, /help       Help for a command")
 		fmt.Fprintln(os.Stderr, "  /? shortcuts    Help for keyboard shortcuts")
@@ -449,9 +445,6 @@ func generateInteractive(cmd *cobra.Command, opts runOptions) error {
 			} else {
 				usageShow()
 			}
-		case strings.HasPrefix(line, "/usage"):
-			fmt.Printf("prompt tokens:     %d\n", sessionPromptTokens)
-			fmt.Printf("completion tokens: %d\n", sessionCompletionTokens)
 		case strings.HasPrefix(line, "/help"), strings.HasPrefix(line, "/?"):
 			args := strings.Fields(line)
 			if len(args) > 1 {
@@ -506,7 +499,7 @@ func generateInteractive(cmd *cobra.Command, opts runOptions) error {
 
 			opts.Messages = append(opts.Messages, newMessage)
 
-			assistant, metrics, err := chat(cmd, opts)
+			assistant, err := chat(cmd, opts)
 			if err != nil {
 				if strings.Contains(err.Error(), "does not support thinking") ||
 					strings.Contains(err.Error(), "invalid think value") {
@@ -516,10 +509,6 @@ func generateInteractive(cmd *cobra.Command, opts runOptions) error {
 				}
 				return err
 			}
-			if metrics != nil {
-				sessionPromptTokens += int64(metrics.PromptEvalCount)
-				sessionCompletionTokens += int64(metrics.EvalCount)
-			}
 			if assistant != nil {
 				opts.Messages = append(opts.Messages, *assistant)
 			}

docs/docs.json

@@ -71,6 +71,10 @@
     {
       "source": "/api",
       "destination": "/api/introduction"
+    },
+    {
+      "source": "/integrations/clawdbot",
+      "destination": "/integrations/openclaw"
     }
   ],
   "navigation": {
@@ -102,8 +106,8 @@
             "group": "Integrations",
             "pages": [
               "/integrations/claude-code",
-              "/integrations/clawdbot",
               "/integrations/cline",
+              "/integrations/openclaw",
               "/integrations/codex",
               "/integrations/droid",
               "/integrations/goose",

docs/gpu.mdx

@@ -10,6 +10,7 @@ Check your compute compatibility to see if your card is supported:
 
 | Compute Capability | Family              | Cards                                                                                                                          |
 | ------------------ | ------------------- | ------------------------------------------------------------------------------------------------------------------------------ |
+| 12.1               | NVIDIA              | `GB10 (DGX Spark)`                                                                                                             |
 | 12.0               | GeForce RTX 50xx    | `RTX 5060` `RTX 5060 Ti` `RTX 5070` `RTX 5070 Ti` `RTX 5080` `RTX 5090`                                                        |
 |                    | NVIDIA Professional | `RTX PRO 4000 Blackwell` `RTX PRO 4500 Blackwell` `RTX PRO 5000 Blackwell` `RTX PRO 6000 Blackwell`                            |
 | 9.0                | NVIDIA              | `H200` `H100`                                                                                                                  |
@@ -163,4 +164,4 @@ To select specific Vulkan GPU(s), you can set the environment variable
 `GGML_VK_VISIBLE_DEVICES` to one or more numeric IDs on the Ollama server as
 described in the [FAQ](faq#how-do-i-configure-ollama-server). If you
 encounter any problems with Vulkan based GPUs, you can disable all Vulkan GPUs
-by setting `GGML_VK_VISIBLE_DEVICES=-1` 
+by setting `GGML_VK_VISIBLE_DEVICES=-1` 

docs/import.mdx

@@ -134,22 +134,12 @@ success
 
 ### Supported Quantizations
 
-- `q4_0`
-- `q4_1`
-- `q5_0`
-- `q5_1`
 - `q8_0`
 
 #### K-means Quantizations
 
-- `q3_K_S`
-- `q3_K_M`
-- `q3_K_L`
 - `q4_K_S`
 - `q4_K_M`
-- `q5_K_S`
-- `q5_K_M`
-- `q6_K`
 
 ## Sharing your model on ollama.com
 

docs/integrations/openclaw.mdx

@@ -1,41 +1,43 @@
 ---
-title: Clawdbot
+title: OpenClaw
 ---
 
-Clawdbot is a personal AI assistant that runs on your own devices. It bridges messaging services (WhatsApp, Telegram, Slack, Discord, iMessage, and more) to AI coding agents through a centralized gateway.
+OpenClaw is a personal AI assistant that runs on your own devices. It bridges messaging services (WhatsApp, Telegram, Slack, Discord, iMessage, and more) to AI coding agents through a centralized gateway.
 
 ## Install
 
-Install [Clawdbot](https://clawd.bot/) 
+Install [OpenClaw](https://openclaw.ai/) 
 
 ```bash
-npm install -g clawdbot@latest
+npm install -g openclaw@latest
 ```
 
 Then run the onboarding wizard:
 
 ```bash
-clawdbot onboard --install-daemon
+openclaw onboard --install-daemon
 ```
 
-<Note>Clawdbot requires a larger context window. It is recommended to use a context window of at least 64k tokens. See [Context length](/context-length) for more information.</Note>
+<Note>OpenClaw requires a larger context window. It is recommended to use a context window of at least 64k tokens. See [Context length](/context-length) for more information.</Note>
 
 ## Usage with Ollama
 
 ### Quick setup
 
 ```bash
-ollama launch clawdbot
+ollama launch openclaw
 ```
 
-This configures Clawdbot to use Ollama and starts the gateway.
+<Note>Previously known as Clawdbot. `ollama launch clawdbot` still works as an alias.</Note>
+
+This configures OpenClaw to use Ollama and starts the gateway.
 If the gateway is already running, no changes need to be made as the gateway will auto-reload the changes. 
 
 
 To configure without launching:
 
 ```shell
-ollama launch clawdbot --config
+ollama launch openclaw --config
 ```
 
 ## Recommended Models

docs/integrations/opencode.mdx

@@ -9,7 +9,7 @@ OpenCode is an open-source AI coding assistant that runs in your terminal.
 Install the [OpenCode CLI](https://opencode.ai):
 
 ```bash
-curl -fsSL https://opencode.ai/install.sh | bash
+curl -fsSL https://opencode.ai/install | bash
 ```
 
 <Note>OpenCode requires a larger context window. It is recommended to use a context window of at least 64k tokens. See [Context length](/context-length) for more information.</Note>

envconfig/config.go

@@ -201,7 +201,7 @@ var (
 	// Enable the new Ollama engine
 	NewEngine = Bool("OLLAMA_NEW_ENGINE")
 	// ContextLength sets the default context length
-	ContextLength = Uint("OLLAMA_CONTEXT_LENGTH", 4096)
+	ContextLength = Uint("OLLAMA_CONTEXT_LENGTH", 0)
 	// Auth enables authentication between the Ollama client and server
 	UseAuth = Bool("OLLAMA_AUTH")
 	// Enable Vulkan backend
@@ -290,7 +290,7 @@ func AsMap() map[string]EnvVar {
 		"OLLAMA_ORIGINS":           {"OLLAMA_ORIGINS", AllowedOrigins(), "A comma separated list of allowed origins"},
 		"OLLAMA_SCHED_SPREAD":      {"OLLAMA_SCHED_SPREAD", SchedSpread(), "Always schedule model across all GPUs"},
 		"OLLAMA_MULTIUSER_CACHE":   {"OLLAMA_MULTIUSER_CACHE", MultiUserCache(), "Optimize prompt caching for multi-user scenarios"},
-		"OLLAMA_CONTEXT_LENGTH":    {"OLLAMA_CONTEXT_LENGTH", ContextLength(), "Context length to use unless otherwise specified (default: 4096)"},
+		"OLLAMA_CONTEXT_LENGTH":    {"OLLAMA_CONTEXT_LENGTH", ContextLength(), "Context length to use unless otherwise specified (default: 4k/32k/256k based on VRAM)"},
 		"OLLAMA_NEW_ENGINE":        {"OLLAMA_NEW_ENGINE", NewEngine(), "Enable the new Ollama engine"},
 		"OLLAMA_REMOTES":           {"OLLAMA_REMOTES", Remotes(), "Allowed hosts for remote models (default \"ollama.com\")"},
 

envconfig/config_test.go

@@ -282,7 +282,7 @@ func TestVar(t *testing.T) {
 
 func TestContextLength(t *testing.T) {
 	cases := map[string]uint{
-		"":     4096,
+		"":     0,
 		"2048": 2048,
 	}
 

llm/server.go

@@ -80,6 +80,7 @@ type LlamaServer interface {
 	GetPort() int
 	GetDeviceInfos(ctx context.Context) []ml.DeviceInfo
 	HasExited() bool
+	ContextLength() int
 }
 
 // llmServer is an instance of a runner hosting a single model
@@ -1200,7 +1201,8 @@ func (s *llmServer) initModel(ctx context.Context, req LoadRequest, operation Lo
 
 	resp, err := http.DefaultClient.Do(r)
 	if err != nil {
-		return nil, fmt.Errorf("do load request: %w", err)
+		slog.Error("do load request", "error", err)
+		return nil, errors.New("model failed to load, this may be due to resource limitations or an internal error, check ollama server logs for details")
 	}
 	defer resp.Body.Close()
 
@@ -1901,6 +1903,10 @@ func (s *llmServer) VRAMByGPU(id ml.DeviceID) uint64 {
 	return 0
 }
 
+func (s *llmServer) ContextLength() int {
+	return s.options.NumCtx
+}
+
 func (s *ollamaServer) GetDeviceInfos(ctx context.Context) []ml.DeviceInfo {
 	devices, err := ml.GetDevicesFromRunner(ctx, s)
 	if err != nil {

runner/ollamarunner/runner.go

@@ -1358,7 +1358,7 @@ func (s *Server) info(w http.ResponseWriter, r *http.Request) {
 		// Dummy load to get the backend wired up
 		f, err := os.CreateTemp("", "*.bin")
 		if err != nil {
-			http.Error(w, fmt.Sprintf("failed to initialize baackend: %v", err), http.StatusInternalServerError)
+			http.Error(w, fmt.Sprintf("failed to initialize backend: %v", err), http.StatusInternalServerError)
 			return
 		}
 		defer f.Close()
@@ -1368,13 +1368,13 @@ func (s *Server) info(w http.ResponseWriter, r *http.Request) {
 			"general.architecture": "llama",
 			"tokenizer.ggml.model": "gpt2",
 		}, nil); err != nil {
-			http.Error(w, fmt.Sprintf("failed to initialize baackend: %v", err), http.StatusInternalServerError)
+			http.Error(w, fmt.Sprintf("failed to initialize backend: %v", err), http.StatusInternalServerError)
 			return
 		}
 
 		m, err = model.New(f.Name(), ml.BackendParams{NumThreads: runtime.NumCPU(), AllocMemory: false, GPULayers: ml.GPULayersList{{}}})
 		if err != nil {
-			http.Error(w, fmt.Sprintf("failed to initialize baackend: %v", err), http.StatusInternalServerError)
+			http.Error(w, fmt.Sprintf("failed to initialize backend: %v", err), http.StatusInternalServerError)
 			return
 		}
 		slog.Debug("dummy model load took", "duration", time.Since(startLoad))

server/routes.go

@@ -75,16 +75,12 @@ func experimentEnabled(name string) bool {
 
 var useClient2 = experimentEnabled("client2")
 
-// Low VRAM mode is based on the sum of total VRAM (not free) and triggers
-// reduced context length on some models
-var lowVRAMThreshold uint64 = 20 * format.GibiByte
-
 var mode string = gin.DebugMode
 
 type Server struct {
-	addr    net.Addr
-	sched   *Scheduler
-	lowVRAM bool
+	addr          net.Addr
+	sched         *Scheduler
+	defaultNumCtx int
 }
 
 func init() {
@@ -107,8 +103,12 @@ var (
 	errBadTemplate = errors.New("template error")
 )
 
-func modelOptions(model *Model, requestOpts map[string]any) (api.Options, error) {
+func (s *Server) modelOptions(model *Model, requestOpts map[string]any) (api.Options, error) {
 	opts := api.DefaultOptions()
+	if opts.NumCtx == 0 {
+		opts.NumCtx = s.defaultNumCtx
+	}
+
 	if err := opts.FromMap(model.Options); err != nil {
 		return api.Options{}, err
 	}
@@ -140,20 +140,11 @@ func (s *Server) scheduleRunner(ctx context.Context, name string, caps []model.C
 		return nil, nil, nil, fmt.Errorf("%s %w", name, err)
 	}
 
-	opts, err := modelOptions(model, requestOpts)
+	opts, err := s.modelOptions(model, requestOpts)
 	if err != nil {
 		return nil, nil, nil, err
 	}
 
-	// This model is much more capable with a larger context, so set that
-	// unless it would penalize performance too much
-	if !s.lowVRAM && slices.Contains([]string{
-		"gptoss", "gpt-oss",
-		"qwen3vl", "qwen3vlmoe",
-	}, model.Config.ModelFamily) {
-		opts.NumCtx = max(opts.NumCtx, 8192)
-	}
-
 	runnerCh, errCh := s.sched.GetRunner(ctx, model, opts, keepAlive)
 	var runner *runnerRef
 	select {
@@ -1720,10 +1711,18 @@ func Serve(ln net.Listener) error {
 	for _, gpu := range gpus {
 		totalVRAM += gpu.TotalMemory - envconfig.GpuOverhead()
 	}
-	if totalVRAM < lowVRAMThreshold {
-		s.lowVRAM = true
-		slog.Info("entering low vram mode", "total vram", format.HumanBytes2(totalVRAM), "threshold", format.HumanBytes2(lowVRAMThreshold))
+
+	// Set default context based on VRAM tier
+	// Use slightly lower thresholds (47/23 GiB vs. 48/24 GiB) to account for small differences in the exact value
+	switch {
+	case totalVRAM >= 47*format.GibiByte:
+		s.defaultNumCtx = 262144
+	case totalVRAM >= 23*format.GibiByte:
+		s.defaultNumCtx = 32768
+	default:
+		s.defaultNumCtx = 4096
 	}
+	slog.Info("vram-based default context", "total_vram", format.HumanBytes2(totalVRAM), "default_num_ctx", s.defaultNumCtx)
 
 	err = srvr.Serve(ln)
 	// If server is closed from the signal handler, wait for the ctx to be done
@@ -1897,8 +1896,8 @@ func (s *Server) PsHandler(c *gin.Context) {
 			Details:   modelDetails,
 			ExpiresAt: v.expiresAt,
 		}
-		if v.Options != nil {
-			mr.ContextLength = v.Options.NumCtx
+		if v.llama != nil {
+			mr.ContextLength = v.llama.ContextLength()
 		}
 		// The scheduler waits to set expiresAt, so if a model is loading it's
 		// possible that it will be set to the unix epoch. For those cases, just

server/routes_debug_test.go

@@ -15,6 +15,7 @@ import (
 )
 
 func TestGenerateDebugRenderOnly(t *testing.T) {
+	t.Setenv("OLLAMA_CONTEXT_LENGTH", "4096")
 	gin.SetMode(gin.TestMode)
 
 	mock := mockRunner{
@@ -208,6 +209,7 @@ func TestGenerateDebugRenderOnly(t *testing.T) {
 }
 
 func TestChatDebugRenderOnly(t *testing.T) {
+	t.Setenv("OLLAMA_CONTEXT_LENGTH", "4096")
 	gin.SetMode(gin.TestMode)
 
 	mock := mockRunner{

server/routes_generate_renderer_test.go

@@ -20,6 +20,7 @@ import (
 // TestGenerateWithBuiltinRenderer tests that api/generate uses built-in renderers
 // when in chat-like flow (messages present, no suffix, no template)
 func TestGenerateWithBuiltinRenderer(t *testing.T) {
+	t.Setenv("OLLAMA_CONTEXT_LENGTH", "4096")
 	gin.SetMode(gin.TestMode)
 
 	mock := mockRunner{
@@ -204,6 +205,7 @@ func TestGenerateWithBuiltinRenderer(t *testing.T) {
 
 // TestGenerateWithDebugRenderOnly tests that debug_render_only works with built-in renderers
 func TestGenerateWithDebugRenderOnly(t *testing.T) {
+	t.Setenv("OLLAMA_CONTEXT_LENGTH", "4096")
 	gin.SetMode(gin.TestMode)
 
 	mock := mockRunner{

server/routes_generate_test.go

@@ -162,6 +162,7 @@ func TestGenerateChatRemote(t *testing.T) {
 }
 
 func TestGenerateChat(t *testing.T) {
+	t.Setenv("OLLAMA_CONTEXT_LENGTH", "4096")
 	gin.SetMode(gin.TestMode)
 
 	mock := mockRunner{
@@ -878,6 +879,7 @@ func TestGenerateChat(t *testing.T) {
 }
 
 func TestGenerate(t *testing.T) {
+	t.Setenv("OLLAMA_CONTEXT_LENGTH", "4096")
 	gin.SetMode(gin.TestMode)
 
 	mock := mockRunner{
@@ -2355,6 +2357,7 @@ func TestGenerateWithImages(t *testing.T) {
 // TestImageGenerateStreamFalse tests that image generation respects stream=false
 // and returns a single JSON response instead of streaming ndjson.
 func TestImageGenerateStreamFalse(t *testing.T) {
+	t.Setenv("OLLAMA_CONTEXT_LENGTH", "4096")
 	gin.SetMode(gin.TestMode)
 
 	p := t.TempDir()

server/routes_options_test.go

@@ -0,0 +1,127 @@
+package server
+
+import (
+	"testing"
+)
+
+func TestModelOptionsNumCtxPriority(t *testing.T) {
+	tests := []struct {
+		name           string
+		envContextLen  string // empty means not set (uses 0 sentinel)
+		defaultNumCtx  int    // VRAM-based default
+		modelNumCtx    int    // 0 means not set in model
+		requestNumCtx  int    // 0 means not set in request
+		expectedNumCtx int
+	}{
+		{
+			name:           "vram default when nothing else set",
+			envContextLen:  "",
+			defaultNumCtx:  32768,
+			modelNumCtx:    0,
+			requestNumCtx:  0,
+			expectedNumCtx: 32768,
+		},
+		{
+			name:           "env var overrides vram default",
+			envContextLen:  "8192",
+			defaultNumCtx:  32768,
+			modelNumCtx:    0,
+			requestNumCtx:  0,
+			expectedNumCtx: 8192,
+		},
+		{
+			name:           "model overrides vram default",
+			envContextLen:  "",
+			defaultNumCtx:  32768,
+			modelNumCtx:    16384,
+			requestNumCtx:  0,
+			expectedNumCtx: 16384,
+		},
+		{
+			name:           "model overrides env var",
+			envContextLen:  "8192",
+			defaultNumCtx:  32768,
+			modelNumCtx:    16384,
+			requestNumCtx:  0,
+			expectedNumCtx: 16384,
+		},
+		{
+			name:           "request overrides everything",
+			envContextLen:  "8192",
+			defaultNumCtx:  32768,
+			modelNumCtx:    16384,
+			requestNumCtx:  4096,
+			expectedNumCtx: 4096,
+		},
+		{
+			name:           "request overrides vram default",
+			envContextLen:  "",
+			defaultNumCtx:  32768,
+			modelNumCtx:    0,
+			requestNumCtx:  4096,
+			expectedNumCtx: 4096,
+		},
+		{
+			name:           "request overrides model",
+			envContextLen:  "",
+			defaultNumCtx:  32768,
+			modelNumCtx:    16384,
+			requestNumCtx:  4096,
+			expectedNumCtx: 4096,
+		},
+		{
+			name:           "low vram tier default",
+			envContextLen:  "",
+			defaultNumCtx:  4096,
+			modelNumCtx:    0,
+			requestNumCtx:  0,
+			expectedNumCtx: 4096,
+		},
+		{
+			name:           "high vram tier default",
+			envContextLen:  "",
+			defaultNumCtx:  262144,
+			modelNumCtx:    0,
+			requestNumCtx:  0,
+			expectedNumCtx: 262144,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			// Set or clear environment variable
+			if tt.envContextLen != "" {
+				t.Setenv("OLLAMA_CONTEXT_LENGTH", tt.envContextLen)
+			}
+
+			// Create server with VRAM-based default
+			s := &Server{
+				defaultNumCtx: tt.defaultNumCtx,
+			}
+
+			// Create model options (use float64 as FromMap expects JSON-style numbers)
+			var modelOpts map[string]any
+			if tt.modelNumCtx != 0 {
+				modelOpts = map[string]any{"num_ctx": float64(tt.modelNumCtx)}
+			}
+			model := &Model{
+				Options: modelOpts,
+			}
+
+			// Create request options (use float64 as FromMap expects JSON-style numbers)
+			var requestOpts map[string]any
+			if tt.requestNumCtx != 0 {
+				requestOpts = map[string]any{"num_ctx": float64(tt.requestNumCtx)}
+			}
+
+			opts, err := s.modelOptions(model, requestOpts)
+			if err != nil {
+				t.Fatalf("modelOptions failed: %v", err)
+			}
+
+			if opts.NumCtx != tt.expectedNumCtx {
+				t.Errorf("NumCtx = %d, want %d", opts.NumCtx, tt.expectedNumCtx)
+			}
+		})
+	}
+}

server/sched_test.go

@@ -804,6 +804,7 @@ func (s *mockLlm) GetPort() int                                       { return -
 func (s *mockLlm) GetDeviceInfos(ctx context.Context) []ml.DeviceInfo { return nil }
 func (s *mockLlm) HasExited() bool                                    { return false }
 func (s *mockLlm) GetActiveDeviceIDs() []ml.DeviceID                  { return nil }
+func (s *mockLlm) ContextLength() int                                 { return 0 }
 
 // TestImageGenRunnerCanBeEvicted verifies that an image generation model
 // loaded in the scheduler can be evicted when idle.

x/imagegen/server.go

@@ -347,6 +347,11 @@ func (s *Server) VRAMByGPU(id ml.DeviceID) uint64 {
 	return s.vramSize
 }
 
+// Context length is not applicable for image generation.
+func (s *Server) ContextLength() int {
+	return 0
+}
+
 func (s *Server) Embedding(ctx context.Context, input string) ([]float32, int, error) {
 	return nil, 0, errors.New("not supported")
 }