fix lint

Add a new /api/usage endpoint that shows aggregate usage statistics per model since the server started.

api/types.go

@@ -922,6 +922,19 @@ type UserResponse struct {
 	Plan      string    `json:"plan,omitempty"`
 }
 
+type UsageResponse struct {
+	// Start is the time the server started tracking usage (UTC, RFC 3339).
+	Start time.Time        `json:"start"`
+	Usage []ModelUsageData `json:"usage"`
+}
+
+type ModelUsageData struct {
+	Model            string `json:"model"`
+	Requests         int64  `json:"requests"`
+	PromptTokens     int64  `json:"prompt_tokens"`
+	CompletionTokens int64  `json:"completion_tokens"`
+}
+
 // Tensor describes the metadata for a given tensor.
 type Tensor struct {
 	Name  string   `json:"name"`

cmd/config/codex.go

@@ -6,7 +6,6 @@ import (
 	"os/exec"
 	"strings"
 
-	"github.com/ollama/ollama/envconfig"
 	"golang.org/x/mod/semver"
 )
 
@@ -33,10 +32,6 @@ func (c *Codex) Run(model string, args []string) error {
 	cmd.Stdin = os.Stdin
 	cmd.Stdout = os.Stdout
 	cmd.Stderr = os.Stderr
-	cmd.Env = append(os.Environ(),
-		"OPENAI_BASE_URL="+envconfig.Host().String()+"/v1/",
-		"OPENAI_API_KEY=ollama",
-	)
 	return cmd.Run()
 }
 

cmd/tui/selector.go

@@ -415,12 +415,6 @@ type multiSelectorModel struct {
 	cancelled    bool
 	confirmed    bool
 	width        int
-
-	// multi enables full multi-select editing mode. The zero value (false)
-	// shows a single-select picker where Enter adds the chosen model to
-	// the existing list. Tab toggles between modes.
-	multi     bool
-	singleAdd string // model picked in single mode
 }
 
 func newMultiSelectorModel(title string, items []SelectItem, preChecked []string) multiSelectorModel {
@@ -435,23 +429,13 @@ func newMultiSelectorModel(title string, items []SelectItem, preChecked []string
 		m.itemIndex[item.Name] = i
 	}
 
-	// Reverse order so preChecked[0] (the current default) ends up last
-	// in checkOrder, matching the "last checked = default" convention.
-	for i := len(preChecked) - 1; i >= 0; i-- {
-		if idx, ok := m.itemIndex[preChecked[i]]; ok {
+	for _, name := range preChecked {
+		if idx, ok := m.itemIndex[name]; ok {
 			m.checked[idx] = true
 			m.checkOrder = append(m.checkOrder, idx)
 		}
 	}
 
-	// Position cursor on the current default model
-	if len(preChecked) > 0 {
-		if idx, ok := m.itemIndex[preChecked[0]]; ok {
-			m.cursor = idx
-			m.updateScroll(m.otherStart())
-		}
-	}
-
 	return m
 }
 
@@ -562,25 +546,14 @@ func (m multiSelectorModel) Update(msg tea.Msg) (tea.Model, tea.Cmd) {
 			m.cancelled = true
 			return m, tea.Quit
 
-		case tea.KeyTab:
-			m.multi = !m.multi
-
 		case tea.KeyEnter:
-			if !m.multi {
-				if len(filtered) > 0 && m.cursor < len(filtered) {
-					m.singleAdd = filtered[m.cursor].Name
-					m.confirmed = true
-					return m, tea.Quit
-				}
-			} else if len(m.checkOrder) > 0 {
+			if len(m.checkOrder) > 0 {
 				m.confirmed = true
 				return m, tea.Quit
 			}
 
 		case tea.KeySpace:
-			if m.multi {
-				m.toggleItem()
-			}
+			m.toggleItem()
 
 		case tea.KeyUp:
 			if m.cursor > 0 {
@@ -619,9 +592,7 @@ func (m multiSelectorModel) Update(msg tea.Msg) (tea.Model, tea.Cmd) {
 			// On some terminals (e.g. Windows PowerShell), space arrives as
 			// KeyRunes instead of KeySpace. Intercept it so toggle still works.
 			if len(msg.Runes) == 1 && msg.Runes[0] == ' ' {
-				if m.multi {
-					m.toggleItem()
-				}
+				m.toggleItem()
 			} else {
 				m.filter += string(msg.Runes)
 				m.cursor = 0
@@ -633,19 +604,6 @@ func (m multiSelectorModel) Update(msg tea.Msg) (tea.Model, tea.Cmd) {
 	return m, nil
 }
 
-func (m multiSelectorModel) renderSingleItem(s *strings.Builder, item SelectItem, idx int) {
-	if idx == m.cursor {
-		s.WriteString(selectorSelectedItemStyle.Render("▸ " + item.Name))
-	} else {
-		s.WriteString(selectorItemStyle.Render(item.Name))
-	}
-	s.WriteString("\n")
-	if item.Description != "" {
-		s.WriteString(selectorDescLineStyle.Render(item.Description))
-		s.WriteString("\n")
-	}
-}
-
 func (m multiSelectorModel) renderMultiItem(s *strings.Builder, item SelectItem, idx int) {
 	origIdx := m.itemIndex[item.Name]
 
@@ -657,7 +615,7 @@ func (m multiSelectorModel) renderMultiItem(s *strings.Builder, item SelectItem,
 	}
 
 	suffix := ""
-	if len(m.checkOrder) > 0 && m.checkOrder[len(m.checkOrder)-1] == origIdx {
+	if len(m.checkOrder) > 0 && m.checkOrder[0] == origIdx {
 		suffix = " " + selectorDefaultTagStyle.Render("(default)")
 	}
 
@@ -679,11 +637,6 @@ func (m multiSelectorModel) View() string {
 		return ""
 	}
 
-	renderItem := m.renderSingleItem
-	if m.multi {
-		renderItem = m.renderMultiItem
-	}
-
 	var s strings.Builder
 
 	s.WriteString(selectorTitleStyle.Render(m.title))
@@ -708,7 +661,7 @@ func (m multiSelectorModel) View() string {
 			if idx >= len(filtered) {
 				break
 			}
-			renderItem(&s, filtered[idx], idx)
+			m.renderMultiItem(&s, filtered[idx], idx)
 		}
 
 		if remaining := len(filtered) - m.scrollOffset - displayCount; remaining > 0 {
@@ -731,7 +684,7 @@ func (m multiSelectorModel) View() string {
 			s.WriteString(sectionHeaderStyle.Render("Recommended"))
 			s.WriteString("\n")
 			for _, idx := range recItems {
-				renderItem(&s, filtered[idx], idx)
+				m.renderMultiItem(&s, filtered[idx], idx)
 			}
 		}
 
@@ -751,7 +704,7 @@ func (m multiSelectorModel) View() string {
 				if idx >= len(otherItems) {
 					break
 				}
-				renderItem(&s, filtered[otherItems[idx]], otherItems[idx])
+				m.renderMultiItem(&s, filtered[otherItems[idx]], otherItems[idx])
 			}
 
 			if remaining := len(otherItems) - m.scrollOffset - displayCount; remaining > 0 {
@@ -763,18 +716,15 @@ func (m multiSelectorModel) View() string {
 
 	s.WriteString("\n")
 
-	if !m.multi {
-		s.WriteString(selectorHelpStyle.Render("↑/↓ navigate • enter select • tab add multiple • esc cancel"))
+	count := m.selectedCount()
+	if count == 0 {
+		s.WriteString(selectorDescStyle.Render("  Select at least one model."))
 	} else {
-		count := m.selectedCount()
-		if count == 0 {
-			s.WriteString(selectorDescStyle.Render("  Select at least one model."))
-		} else {
-			s.WriteString(selectorDescStyle.Render(fmt.Sprintf("  %d selected - press enter to continue", count)))
-		}
-		s.WriteString("\n\n")
-		s.WriteString(selectorHelpStyle.Render("↑/↓ navigate • space toggle • tab select single • enter confirm • esc cancel"))
+		s.WriteString(selectorDescStyle.Render(fmt.Sprintf("  %d selected - press enter to continue", count)))
 	}
+	s.WriteString("\n\n")
+
+	s.WriteString(selectorHelpStyle.Render("↑/↓ navigate • space toggle • enter confirm • esc cancel"))
 
 	result := s.String()
 	if m.width > 0 {
@@ -797,28 +747,18 @@ func SelectMultiple(title string, items []SelectItem, preChecked []string) ([]st
 	}
 
 	fm := finalModel.(multiSelectorModel)
-	if fm.cancelled || !fm.confirmed {
+	if fm.cancelled {
 		return nil, ErrCancelled
 	}
 
-	// Single-add mode: prepend the picked model, keep existing models deduped
-	if fm.singleAdd != "" {
-		result := []string{fm.singleAdd}
-		for _, name := range preChecked {
-			if name != fm.singleAdd {
-				result = append(result, name)
-			}
-		}
-		return result, nil
+	if !fm.confirmed {
+		return nil, ErrCancelled
 	}
 
-	// Multi-edit mode: last checked is default (first in result)
-	last := fm.checkOrder[len(fm.checkOrder)-1]
-	result := []string{fm.items[last].Name}
+	var result []string
 	for _, idx := range fm.checkOrder {
-		if idx != last {
-			result = append(result, fm.items[idx].Name)
-		}
+		result = append(result, fm.items[idx].Name)
 	}
+
 	return result, nil
 }

cmd/tui/selector_test.go

@@ -539,7 +539,6 @@ func TestMultiView_CursorIndicator(t *testing.T) {
 
 func TestMultiView_CheckedItemShowsX(t *testing.T) {
 	m := newMultiSelectorModel("Pick:", items("a", "b"), []string{"a"})
-	m.multi = true
 	content := m.View()
 
 	if !strings.Contains(content, "[x]") {
@@ -551,18 +550,11 @@ func TestMultiView_CheckedItemShowsX(t *testing.T) {
 }
 
 func TestMultiView_DefaultTag(t *testing.T) {
-	m := newMultiSelectorModel("Pick:", items("a", "b", "c"), []string{"a", "b"})
-	m.multi = true
+	m := newMultiSelectorModel("Pick:", items("a", "b"), []string{"a"})
 	content := m.View()
 
 	if !strings.Contains(content, "(default)") {
-		t.Error("should have (default) tag")
-	}
-	// preChecked[0] ("a") should be the default (last in checkOrder)
-	aIdx := strings.Index(content, "a")
-	defaultIdx := strings.Index(content, "(default)")
-	if defaultIdx < aIdx {
-		t.Error("(default) tag should appear after 'a' (the current default)")
+		t.Error("first checked item should have (default) tag")
 	}
 }
 
@@ -593,7 +585,6 @@ func TestMultiView_OverflowIndicator(t *testing.T) {
 
 func TestMultiUpdate_SpaceTogglesItem(t *testing.T) {
 	m := newMultiSelectorModel("Pick:", items("a", "b", "c"), nil)
-	m.multi = true
 	m.cursor = 1
 
 	// Simulate space delivered as tea.KeySpace
@@ -610,7 +601,6 @@ func TestMultiUpdate_SpaceTogglesItem(t *testing.T) {
 
 func TestMultiUpdate_SpaceRuneTogglesItem(t *testing.T) {
 	m := newMultiSelectorModel("Pick:", items("a", "b", "c"), nil)
-	m.multi = true
 	m.cursor = 1
 
 	// Simulate space delivered as tea.KeyRunes (Windows PowerShell behavior)
@@ -628,161 +618,6 @@ func TestMultiUpdate_SpaceRuneTogglesItem(t *testing.T) {
 	}
 }
 
-// --- Single-add mode ---
-
-func TestMulti_StartsInSingleMode(t *testing.T) {
-	m := newMultiSelectorModel("Pick:", items("a", "b"), nil)
-	if m.multi {
-		t.Error("should start in single mode (multi=false)")
-	}
-}
-
-func TestMulti_SingleModeNoCheckboxes(t *testing.T) {
-	m := newMultiSelectorModel("Pick:", items("a", "b"), nil)
-	content := m.View()
-	if strings.Contains(content, "[x]") || strings.Contains(content, "[ ]") {
-		t.Error("single mode should not show checkboxes")
-	}
-	if !strings.Contains(content, "▸") {
-		t.Error("single mode should show cursor indicator")
-	}
-}
-
-func TestMulti_SingleModeEnterPicksItem(t *testing.T) {
-	m := newMultiSelectorModel("Pick:", items("a", "b", "c"), nil)
-	m.cursor = 1
-
-	updated, _ := m.Update(tea.KeyMsg{Type: tea.KeyEnter})
-	m = updated.(multiSelectorModel)
-
-	if m.singleAdd != "b" {
-		t.Errorf("enter in single mode should pick cursor item, got %q", m.singleAdd)
-	}
-	if !m.confirmed {
-		t.Error("should set confirmed")
-	}
-}
-
-func TestMulti_SingleModeSpaceIsNoop(t *testing.T) {
-	m := newMultiSelectorModel("Pick:", items("a", "b"), nil)
-	m.cursor = 0
-
-	updated, _ := m.Update(tea.KeyMsg{Type: tea.KeySpace})
-	m = updated.(multiSelectorModel)
-
-	if len(m.checked) != 0 {
-		t.Error("space in single mode should not toggle items")
-	}
-}
-
-func TestMulti_SingleModeSpaceRuneIsNoop(t *testing.T) {
-	m := newMultiSelectorModel("Pick:", items("a", "b"), nil)
-	m.cursor = 0
-
-	updated, _ := m.Update(tea.KeyMsg{Type: tea.KeyRunes, Runes: []rune{' '}})
-	m = updated.(multiSelectorModel)
-
-	if len(m.checked) != 0 {
-		t.Error("space rune in single mode should not toggle items")
-	}
-	if m.filter != "" {
-		t.Error("space rune in single mode should not add to filter")
-	}
-}
-
-func TestMulti_TabTogglesMode(t *testing.T) {
-	m := newMultiSelectorModel("Pick:", items("a", "b"), nil)
-
-	if m.multi {
-		t.Fatal("should start in single mode")
-	}
-
-	updated, _ := m.Update(tea.KeyMsg{Type: tea.KeyTab})
-	m = updated.(multiSelectorModel)
-	if !m.multi {
-		t.Error("tab should switch to multi mode")
-	}
-
-	updated, _ = m.Update(tea.KeyMsg{Type: tea.KeyTab})
-	m = updated.(multiSelectorModel)
-	if m.multi {
-		t.Error("tab should switch back to single mode")
-	}
-}
-
-func TestMulti_SingleModeHelpText(t *testing.T) {
-	m := newMultiSelectorModel("Pick:", items("a"), nil)
-	content := m.View()
-	if !strings.Contains(content, "tab add multiple") {
-		t.Error("single mode should show 'tab add multiple' in help")
-	}
-}
-
-func TestMulti_MultiModeHelpText(t *testing.T) {
-	m := newMultiSelectorModel("Pick:", items("a"), nil)
-	m.multi = true
-	content := m.View()
-	if !strings.Contains(content, "tab select single") {
-		t.Error("multi mode should show 'tab select single' in help")
-	}
-}
-
-// --- preChecked initialization order ---
-
-func TestMulti_PreCheckedDefaultIsLast(t *testing.T) {
-	// preChecked[0] ("a") is the current default and should end up
-	// last in checkOrder so it gets the (default) tag.
-	m := newMultiSelectorModel("Pick:", items("a", "b", "c"), []string{"a", "b", "c"})
-
-	if len(m.checkOrder) != 3 {
-		t.Fatalf("expected 3 in checkOrder, got %d", len(m.checkOrder))
-	}
-	lastIdx := m.checkOrder[len(m.checkOrder)-1]
-	if m.items[lastIdx].Name != "a" {
-		t.Errorf("preChecked[0] should be last in checkOrder, got %q", m.items[lastIdx].Name)
-	}
-}
-
-func TestMulti_CursorOnDefaultModel(t *testing.T) {
-	// preChecked[0] ("b") is the default; cursor should start on it
-	m := newMultiSelectorModel("Pick:", items("a", "b", "c"), []string{"b", "c"})
-
-	if m.cursor != 1 {
-		t.Errorf("cursor should be on preChecked[0] ('b') at index 1, got %d", m.cursor)
-	}
-}
-
-// --- Multi-mode last-checked is default ---
-
-func TestMulti_LastCheckedIsDefault(t *testing.T) {
-	m := newMultiSelectorModel("Pick:", items("alpha", "beta", "gamma"), nil)
-	m.multi = true
-
-	// Check "alpha" then "gamma"
-	m.cursor = 0
-	m.toggleItem()
-	m.cursor = 2
-	m.toggleItem()
-
-	// Last checked ("gamma") should be at the end of checkOrder
-	lastIdx := m.checkOrder[len(m.checkOrder)-1]
-	if m.items[lastIdx].Name != "gamma" {
-		t.Errorf("last checked should be 'gamma', got %q", m.items[lastIdx].Name)
-	}
-
-	// The (default) tag renders based on checkOrder[len-1]
-	content := m.View()
-	if !strings.Contains(content, "(default)") {
-		t.Fatal("should show (default) tag")
-	}
-	// "alpha" line should NOT have the default tag
-	for _, line := range strings.Split(content, "\n") {
-		if strings.Contains(line, "alpha") && strings.Contains(line, "(default)") {
-			t.Error("'alpha' (first checked) should not have (default) tag")
-		}
-	}
-}
-
 // Key message helpers for testing
 
 type keyType = int

cmd/tui/tui.go

@@ -429,24 +429,8 @@ func (m model) Update(msg tea.Msg) (tea.Model, tea.Cmd) {
 			}
 			if m.multiModalSelector.confirmed {
 				var selected []string
-				if m.multiModalSelector.singleAdd != "" {
-					// Single-add mode: prepend picked model, keep existing deduped
-					selected = []string{m.multiModalSelector.singleAdd}
-					for _, name := range config.IntegrationModels(m.items[m.cursor].integration) {
-						if name != m.multiModalSelector.singleAdd {
-							selected = append(selected, name)
-						}
-					}
-				} else {
-					// Last checked is default (first in result)
-					co := m.multiModalSelector.checkOrder
-					last := co[len(co)-1]
-					selected = []string{m.multiModalSelector.items[last].Name}
-					for _, idx := range co {
-						if idx != last {
-							selected = append(selected, m.multiModalSelector.items[idx].Name)
-						}
-					}
+				for _, idx := range m.multiModalSelector.checkOrder {
+					selected = append(selected, m.multiModalSelector.items[idx].Name)
 				}
 				if len(selected) > 0 {
 					m.changeModels = selected

docs/api.md

@@ -15,6 +15,7 @@
 - [Push a Model](#push-a-model)
 - [Generate Embeddings](#generate-embeddings)
 - [List Running Models](#list-running-models)
+- [Usage](#usage)
 - [Version](#version)
 - [Experimental: Image Generation](#image-generation-experimental)
 
@@ -1854,6 +1855,53 @@ curl http://localhost:11434/api/embeddings -d '{
 }
 ```
 
+## Usage
+
+```
+GET /api/usage
+```
+
+Show aggregate usage statistics per model since the server started. All timestamps are UTC in RFC 3339 format.
+
+### Examples
+
+#### Request
+
+```shell
+curl http://localhost:11434/api/usage
+```
+
+#### Response
+
+```json
+{
+  "start": "2025-01-27T20:00:00Z",
+  "usage": [
+    {
+      "model": "llama3.2",
+      "requests": 5,
+      "prompt_tokens": 130,
+      "completion_tokens": 890
+    },
+    {
+      "model": "deepseek-r1",
+      "requests": 2,
+      "prompt_tokens": 48,
+      "completion_tokens": 312
+    }
+  ]
+}
+```
+
+#### Response fields
+
+- `start`: when the server started tracking usage (UTC, RFC 3339)
+- `usage`: list of per-model usage statistics
+  - `model`: model name
+  - `requests`: total number of completed requests
+  - `prompt_tokens`: total prompt tokens evaluated
+  - `completion_tokens`: total completion tokens generated
+
 ## Version
 
 ```

go.mod

@@ -26,7 +26,6 @@ require (
 	github.com/d4l3k/go-bfloat16 v0.0.0-20211005043715-690c3bdd05f1
 	github.com/dlclark/regexp2 v1.11.4
 	github.com/emirpasic/gods/v2 v2.0.0-alpha
-	github.com/klauspost/compress v1.18.3
 	github.com/mattn/go-runewidth v0.0.16
 	github.com/nlpodyssey/gopickle v0.3.0
 	github.com/pdevine/tensor v0.0.0-20240510204454-f88f4562727c

go.sum

@@ -122,6 +122,7 @@ github.com/golang/protobuf v1.5.0/go.mod h1:FsONVRAS9T7sI+LIUmWTfcYkHO4aIWwzhcaS
 github.com/golang/protobuf v1.5.2/go.mod h1:XVQd3VNwM+JqD3oG2Ue2ip4fOMUkwXdXDdiuN0vRsmY=
 github.com/golang/protobuf v1.5.4 h1:i7eJL8qZTpSEXOPTxNKhASYpMn+8e5Q6AdndVa1dWek=
 github.com/golang/protobuf v1.5.4/go.mod h1:lnTiLA8Wa4RWRcIUkrtSVa5nRhsEGBg48fD6rSs7xps=
+github.com/golang/snappy v0.0.3 h1:fHPg5GQYlCeLIPB9BZqMVR5nR9A+IM5zcgeTdjMYmLA=
 github.com/golang/snappy v0.0.3/go.mod h1:/XxbfmMg8lxefKM7IXC3fBNl/7bRcc72aCRzEWrmP2Q=
 github.com/google/flatbuffers v2.0.0+incompatible/go.mod h1:1AeVuKshWv4vARoZatz6mlQ0JxURH0Kv5+zNeJKJCa8=
 github.com/google/flatbuffers v24.3.25+incompatible h1:CX395cjN9Kke9mmalRoL3d81AtFUxJM+yDthflgJGkI=
@@ -149,9 +150,8 @@ github.com/jung-kurt/gofpdf v1.0.0/go.mod h1:7Id9E/uU8ce6rXgefFLlgrJj/GYY22cpxn+
 github.com/jung-kurt/gofpdf v1.0.3-0.20190309125859-24315acbbda5/go.mod h1:7Id9E/uU8ce6rXgefFLlgrJj/GYY22cpxn+r32jIOes=
 github.com/kisielk/errcheck v1.5.0/go.mod h1:pFxgyoBC7bSaBwPgfKdkLd5X25qrDl4LWUI2bnpBCr8=
 github.com/kisielk/gotool v1.0.0/go.mod h1:XhKaO+MFFWcvkIS/tQcRk01m1F5IRFswLeQ+oQHNcck=
+github.com/klauspost/compress v1.13.1 h1:wXr2uRxZTJXHLly6qhJabee5JqIhTRoLBhDOA74hDEQ=
 github.com/klauspost/compress v1.13.1/go.mod h1:8dP1Hq4DHOhN9w426knH3Rhby4rFm6D8eO+e+Dq5Gzg=
-github.com/klauspost/compress v1.18.3 h1:9PJRvfbmTabkOX8moIpXPbMMbYN60bWImDDU7L+/6zw=
-github.com/klauspost/compress v1.18.3/go.mod h1:R0h/fSBs8DE4ENlcrlib3PsXS61voFxhIs2DeRhCvJ4=
 github.com/klauspost/cpuid/v2 v2.0.9/go.mod h1:FInQzS24/EEf25PyTYn52gqo7WaD8xa0213Md/qVLRg=
 github.com/klauspost/cpuid/v2 v2.2.7 h1:ZWSB3igEs+d0qvnxR/ZBzXVmxkgt8DdzP6m9pfuVLDM=
 github.com/klauspost/cpuid/v2 v2.2.7/go.mod h1:Lcz8mBdAVJIBVzewtcLocK12l3Y+JytZYpaMropDUws=

middleware/openai.go

@@ -11,7 +11,6 @@ import (
 	"time"
 
 	"github.com/gin-gonic/gin"
-	"github.com/klauspost/compress/zstd"
 
 	"github.com/ollama/ollama/api"
 	"github.com/ollama/ollama/openai"
@@ -497,17 +496,6 @@ func (w *ResponsesWriter) Write(data []byte) (int, error) {
 
 func ResponsesMiddleware() gin.HandlerFunc {
 	return func(c *gin.Context) {
-		if c.GetHeader("Content-Encoding") == "zstd" {
-			reader, err := zstd.NewReader(c.Request.Body, zstd.WithDecoderMaxMemory(8<<20))
-			if err != nil {
-				c.AbortWithStatusJSON(http.StatusBadRequest, openai.NewError(http.StatusBadRequest, "failed to decompress zstd body"))
-				return
-			}
-			defer reader.Close()
-			c.Request.Body = io.NopCloser(reader)
-			c.Request.Header.Del("Content-Encoding")
-		}
-
 		var req openai.ResponsesRequest
 		if err := c.ShouldBindJSON(&req); err != nil {
 			c.AbortWithStatusJSON(http.StatusBadRequest, openai.NewError(http.StatusBadRequest, err.Error()))

middleware/openai_test.go

@@ -14,7 +14,6 @@ import (
 
 	"github.com/gin-gonic/gin"
 	"github.com/google/go-cmp/cmp"
-	"github.com/klauspost/compress/zstd"
 
 	"github.com/ollama/ollama/api"
 	"github.com/ollama/ollama/openai"
@@ -1239,102 +1238,3 @@ func TestImageEditsMiddleware(t *testing.T) {
 		})
 	}
 }
-
-func zstdCompress(t *testing.T, data []byte) []byte {
-	t.Helper()
-	var buf bytes.Buffer
-	w, err := zstd.NewWriter(&buf)
-	if err != nil {
-		t.Fatal(err)
-	}
-	if _, err := w.Write(data); err != nil {
-		t.Fatal(err)
-	}
-	if err := w.Close(); err != nil {
-		t.Fatal(err)
-	}
-	return buf.Bytes()
-}
-
-func TestResponsesMiddlewareZstd(t *testing.T) {
-	tests := []struct {
-		name        string
-		body        string
-		useZstd     bool
-		oversized   bool
-		wantCode    int
-		wantModel   string
-		wantMessage string
-	}{
-		{
-			name:        "plain JSON",
-			body:        `{"model": "test-model", "input": "Hello"}`,
-			wantCode:    http.StatusOK,
-			wantModel:   "test-model",
-			wantMessage: "Hello",
-		},
-		{
-			name:        "zstd compressed",
-			body:        `{"model": "test-model", "input": "Hello"}`,
-			useZstd:     true,
-			wantCode:    http.StatusOK,
-			wantModel:   "test-model",
-			wantMessage: "Hello",
-		},
-		{
-			name:      "zstd over max decompressed size",
-			oversized: true,
-			useZstd:   true,
-			wantCode:  http.StatusBadRequest,
-		},
-	}
-
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			var capturedRequest *api.ChatRequest
-
-			gin.SetMode(gin.TestMode)
-			router := gin.New()
-			router.Use(ResponsesMiddleware(), captureRequestMiddleware(&capturedRequest))
-			router.Handle(http.MethodPost, "/v1/responses", func(c *gin.Context) {
-				c.Status(http.StatusOK)
-			})
-
-			var bodyReader io.Reader
-			if tt.oversized {
-				bodyReader = bytes.NewReader(zstdCompress(t, bytes.Repeat([]byte("A"), 9<<20)))
-			} else if tt.useZstd {
-				bodyReader = bytes.NewReader(zstdCompress(t, []byte(tt.body)))
-			} else {
-				bodyReader = strings.NewReader(tt.body)
-			}
-
-			req, _ := http.NewRequest(http.MethodPost, "/v1/responses", bodyReader)
-			req.Header.Set("Content-Type", "application/json")
-			if tt.useZstd || tt.oversized {
-				req.Header.Set("Content-Encoding", "zstd")
-			}
-
-			resp := httptest.NewRecorder()
-			router.ServeHTTP(resp, req)
-
-			if resp.Code != tt.wantCode {
-				t.Fatalf("expected status %d, got %d: %s", tt.wantCode, resp.Code, resp.Body.String())
-			}
-
-			if tt.wantCode != http.StatusOK {
-				return
-			}
-
-			if capturedRequest == nil {
-				t.Fatal("expected captured request, got nil")
-			}
-			if capturedRequest.Model != tt.wantModel {
-				t.Fatalf("expected model %q, got %q", tt.wantModel, capturedRequest.Model)
-			}
-			if len(capturedRequest.Messages) != 1 || capturedRequest.Messages[0].Content != tt.wantMessage {
-				t.Fatalf("expected single user message %q, got %+v", tt.wantMessage, capturedRequest.Messages)
-			}
-		})
-	}
-}

scripts/install.sh

@@ -2,10 +2,6 @@
 # This script installs Ollama on Linux and macOS.
 # It detects the current operating system architecture and installs the appropriate version of Ollama.
 
-# Wrap script in main function so that a truncated partial download doesn't end
-# up executing half a script.
-main() {
-
 set -eu
 
 red="$( (/usr/bin/tput bold || :; /usr/bin/tput setaf 1 || :) 2>&-)"
@@ -450,6 +446,3 @@ fi
 
 status "NVIDIA GPU ready."
 install_success
-}
-
-main

server/routes.go

@@ -91,6 +91,8 @@ type Server struct {
 	aliasesOnce   sync.Once
 	aliases       *store
 	aliasesErr    error
+	lowVRAM       bool
+	usage         *UsageTracker
 }
 
 func init() {
@@ -289,6 +291,10 @@ func (s *Server) GenerateHandler(c *gin.Context) {
 		c.Header("Content-Type", contentType)
 
 		fn := func(resp api.GenerateResponse) error {
+			if resp.Done {
+				s.usage.Record(origModel, resp.PromptEvalCount, resp.EvalCount)
+			}
+
 			resp.Model = origModel
 			resp.RemoteModel = m.Config.RemoteModel
 			resp.RemoteHost = m.Config.RemoteHost
@@ -595,6 +601,8 @@ func (s *Server) GenerateHandler(c *gin.Context) {
 					}
 					res.Context = tokens
 				}
+
+				s.usage.Record(req.Model, cr.PromptEvalCount, cr.EvalCount)
 			}
 
 			if builtinParser != nil {
@@ -1622,6 +1630,8 @@ func (s *Server) GenerateRoutes(rc *ollama.Registry) (http.Handler, error) {
 	r.POST("/api/experimental/aliases", s.CreateAliasHandler)
 	r.DELETE("/api/experimental/aliases", s.DeleteAliasHandler)
 
+	r.GET("/api/usage", s.UsageHandler)
+
 	// Inference
 	r.GET("/api/ps", s.PsHandler)
 	r.POST("/api/generate", s.GenerateHandler)
@@ -1692,7 +1702,7 @@ func Serve(ln net.Listener) error {
 		}
 	}
 
-	s := &Server{addr: ln.Addr()}
+	s := &Server{addr: ln.Addr(), usage: NewUsageTracker()}
 
 	var rc *ollama.Registry
 	if useClient2 {
@@ -1927,6 +1937,10 @@ func (s *Server) SignoutHandler(c *gin.Context) {
 	c.JSON(http.StatusOK, nil)
 }
 
+func (s *Server) UsageHandler(c *gin.Context) {
+	c.JSON(http.StatusOK, s.usage.Stats())
+}
+
 func (s *Server) PsHandler(c *gin.Context) {
 	models := []api.ProcessModelResponse{}
 
@@ -2097,6 +2111,10 @@ func (s *Server) ChatHandler(c *gin.Context) {
 		c.Header("Content-Type", contentType)
 
 		fn := func(resp api.ChatResponse) error {
+			if resp.Done {
+				s.usage.Record(origModel, resp.PromptEvalCount, resp.EvalCount)
+			}
+
 			resp.Model = origModel
 			resp.RemoteModel = m.Config.RemoteModel
 			resp.RemoteHost = m.Config.RemoteHost
@@ -2317,6 +2335,8 @@ func (s *Server) ChatHandler(c *gin.Context) {
 					res.DoneReason = r.DoneReason.String()
 					res.TotalDuration = time.Since(checkpointStart)
 					res.LoadDuration = checkpointLoaded.Sub(checkpointStart)
+
+					s.usage.Record(req.Model, r.PromptEvalCount, r.EvalCount)
 				}
 
 				if builtinParser != nil {

server/routes_debug_test.go

@@ -30,6 +30,7 @@ func TestGenerateDebugRenderOnly(t *testing.T) {
 	}
 
 	s := Server{
+		usage: NewUsageTracker(),
 		sched: &Scheduler{
 			pendingReqCh:    make(chan *LlmRequest, 1),
 			finishedReqCh:   make(chan *LlmRequest, 1),
@@ -224,6 +225,7 @@ func TestChatDebugRenderOnly(t *testing.T) {
 	}
 
 	s := Server{
+		usage: NewUsageTracker(),
 		sched: &Scheduler{
 			pendingReqCh:    make(chan *LlmRequest, 1),
 			finishedReqCh:   make(chan *LlmRequest, 1),

server/routes_generate_renderer_test.go

@@ -35,6 +35,7 @@ func TestGenerateWithBuiltinRenderer(t *testing.T) {
 	}
 
 	s := Server{
+		usage: NewUsageTracker(),
 		sched: &Scheduler{
 			pendingReqCh:    make(chan *LlmRequest, 1),
 			finishedReqCh:   make(chan *LlmRequest, 1),
@@ -220,6 +221,7 @@ func TestGenerateWithDebugRenderOnly(t *testing.T) {
 	}
 
 	s := Server{
+		usage: NewUsageTracker(),
 		sched: &Scheduler{
 			pendingReqCh:    make(chan *LlmRequest, 1),
 			finishedReqCh:   make(chan *LlmRequest, 1),

server/routes_generate_test.go

@@ -88,19 +88,39 @@ func TestGenerateChatRemote(t *testing.T) {
 		if r.Method != http.MethodPost {
 			t.Errorf("Expected POST request, got %s", r.Method)
 		}
-		if r.URL.Path != "/api/chat" {
-			t.Errorf("Expected path '/api/chat', got %s", r.URL.Path)
-		}
 
 		w.WriteHeader(http.StatusOK)
 		w.Header().Set("Content-Type", "application/json")
-		resp := api.ChatResponse{
-			Model:      "test",
-			Done:       true,
-			DoneReason: "load",
-		}
-		if err := json.NewEncoder(w).Encode(&resp); err != nil {
-			t.Fatal(err)
+
+		switch r.URL.Path {
+		case "/api/chat":
+			resp := api.ChatResponse{
+				Model:      "test",
+				Done:       true,
+				DoneReason: "load",
+				Metrics: api.Metrics{
+					PromptEvalCount: 10,
+					EvalCount:       20,
+				},
+			}
+			if err := json.NewEncoder(w).Encode(&resp); err != nil {
+				t.Fatal(err)
+			}
+		case "/api/generate":
+			resp := api.GenerateResponse{
+				Model:      "test",
+				Done:       true,
+				DoneReason: "stop",
+				Metrics: api.Metrics{
+					PromptEvalCount: 5,
+					EvalCount:       15,
+				},
+			}
+			if err := json.NewEncoder(w).Encode(&resp); err != nil {
+				t.Fatal(err)
+			}
+		default:
+			t.Errorf("unexpected path %s", r.URL.Path)
 		}
 	}))
 	defer rs.Close()
@@ -111,7 +131,7 @@ func TestGenerateChatRemote(t *testing.T) {
 	}
 
 	t.Setenv("OLLAMA_REMOTES", p.Hostname())
-	s := Server{}
+	s := Server{usage: NewUsageTracker()}
 	w := createRequest(t, s.CreateHandler, api.CreateRequest{
 		Model:      "test-cloud",
 		RemoteHost: rs.URL,
@@ -159,6 +179,61 @@ func TestGenerateChatRemote(t *testing.T) {
 			t.Errorf("expected done reason load, got %s", actual.DoneReason)
 		}
 	})
+
+	t.Run("remote chat usage tracking", func(t *testing.T) {
+		stats := s.usage.Stats()
+		found := false
+		for _, m := range stats.Usage {
+			if m.Model == "test-cloud" {
+				found = true
+				if m.Requests != 1 {
+					t.Errorf("expected 1 request, got %d", m.Requests)
+				}
+				if m.PromptTokens != 10 {
+					t.Errorf("expected 10 prompt tokens, got %d", m.PromptTokens)
+				}
+				if m.CompletionTokens != 20 {
+					t.Errorf("expected 20 completion tokens, got %d", m.CompletionTokens)
+				}
+			}
+		}
+		if !found {
+			t.Error("expected usage entry for test-cloud")
+		}
+	})
+
+	t.Run("remote generate usage tracking", func(t *testing.T) {
+		// Reset the tracker for a clean test
+		s.usage = NewUsageTracker()
+
+		w := createRequest(t, s.GenerateHandler, api.GenerateRequest{
+			Model:  "test-cloud",
+			Prompt: "hello",
+		})
+		if w.Code != http.StatusOK {
+			t.Fatalf("expected status 200, got %d", w.Code)
+		}
+
+		stats := s.usage.Stats()
+		found := false
+		for _, m := range stats.Usage {
+			if m.Model == "test-cloud" {
+				found = true
+				if m.Requests != 1 {
+					t.Errorf("expected 1 request, got %d", m.Requests)
+				}
+				if m.PromptTokens != 5 {
+					t.Errorf("expected 5 prompt tokens, got %d", m.PromptTokens)
+				}
+				if m.CompletionTokens != 15 {
+					t.Errorf("expected 15 completion tokens, got %d", m.CompletionTokens)
+				}
+			}
+		}
+		if !found {
+			t.Error("expected usage entry for test-cloud")
+		}
+	})
 }
 
 func TestGenerateChat(t *testing.T) {
@@ -177,6 +252,7 @@ func TestGenerateChat(t *testing.T) {
 	}
 
 	s := Server{
+		usage: NewUsageTracker(),
 		sched: &Scheduler{
 			pendingReqCh:    make(chan *LlmRequest, 1),
 			finishedReqCh:   make(chan *LlmRequest, 1),
@@ -894,6 +970,7 @@ func TestGenerate(t *testing.T) {
 	}
 
 	s := Server{
+		usage: NewUsageTracker(),
 		sched: &Scheduler{
 			pendingReqCh:    make(chan *LlmRequest, 1),
 			finishedReqCh:   make(chan *LlmRequest, 1),
@@ -1378,6 +1455,7 @@ func TestGenerateLogprobs(t *testing.T) {
 		}
 
 		s := &Server{
+			usage: NewUsageTracker(),
 			sched: &Scheduler{
 				pendingReqCh:    make(chan *LlmRequest, 1),
 				finishedReqCh:   make(chan *LlmRequest, 1),
@@ -1558,6 +1636,7 @@ func TestChatLogprobs(t *testing.T) {
 		}
 
 		s := &Server{
+			usage: NewUsageTracker(),
 			sched: &Scheduler{
 				pendingReqCh:    make(chan *LlmRequest, 1),
 				finishedReqCh:   make(chan *LlmRequest, 1),
@@ -1668,6 +1747,7 @@ func TestChatWithPromptEndingInThinkTag(t *testing.T) {
 		}
 
 		s := &Server{
+			usage: NewUsageTracker(),
 			sched: &Scheduler{
 				pendingReqCh:    make(chan *LlmRequest, 1),
 				finishedReqCh:   make(chan *LlmRequest, 1),
@@ -2114,6 +2194,7 @@ func TestGenerateUnload(t *testing.T) {
 	var loadFnCalled bool
 
 	s := Server{
+		usage: NewUsageTracker(),
 		sched: &Scheduler{
 			pendingReqCh:    make(chan *LlmRequest, 1),
 			finishedReqCh:   make(chan *LlmRequest, 1),
@@ -2215,6 +2296,7 @@ func TestGenerateWithImages(t *testing.T) {
 	}
 
 	s := Server{
+		usage: NewUsageTracker(),
 		sched: &Scheduler{
 			pendingReqCh:    make(chan *LlmRequest, 1),
 			finishedReqCh:   make(chan *LlmRequest, 1),
@@ -2393,6 +2475,7 @@ func TestImageGenerateStreamFalse(t *testing.T) {
 
 	opts := api.DefaultOptions()
 	s := Server{
+		usage: NewUsageTracker(),
 		sched: &Scheduler{
 			pendingReqCh:  make(chan *LlmRequest, 1),
 			finishedReqCh: make(chan *LlmRequest, 1),

server/routes_harmony_streaming_test.go

@@ -255,6 +255,7 @@ func TestChatHarmonyParserStreamingRealtime(t *testing.T) {
 			}
 
 			s := Server{
+				usage: NewUsageTracker(),
 				sched: &Scheduler{
 					pendingReqCh:    make(chan *LlmRequest, 1),
 					finishedReqCh:   make(chan *LlmRequest, 1),
@@ -406,6 +407,7 @@ func TestChatHarmonyParserStreamingSimple(t *testing.T) {
 	}
 
 	s := Server{
+		usage: NewUsageTracker(),
 		sched: &Scheduler{
 			pendingReqCh:    make(chan *LlmRequest, 1),
 			finishedReqCh:   make(chan *LlmRequest, 1),
@@ -588,6 +590,7 @@ func TestChatHarmonyParserStreaming(t *testing.T) {
 			}
 
 			s := Server{
+				usage: NewUsageTracker(),
 				sched: &Scheduler{
 					pendingReqCh:    make(chan *LlmRequest, 1),
 					finishedReqCh:   make(chan *LlmRequest, 1),

server/usage.go

@@ -0,0 +1,62 @@
+package server
+
+import (
+	"sync"
+	"time"
+
+	"github.com/ollama/ollama/api"
+)
+
+type ModelUsage struct {
+	Requests         int64
+	PromptTokens     int64
+	CompletionTokens int64
+}
+
+type UsageTracker struct {
+	mu     sync.Mutex
+	start  time.Time
+	models map[string]*ModelUsage
+}
+
+func NewUsageTracker() *UsageTracker {
+	return &UsageTracker{
+		start:  time.Now().UTC(),
+		models: make(map[string]*ModelUsage),
+	}
+}
+
+func (u *UsageTracker) Record(model string, promptTokens, completionTokens int) {
+	u.mu.Lock()
+	defer u.mu.Unlock()
+
+	m, ok := u.models[model]
+	if !ok {
+		m = &ModelUsage{}
+		u.models[model] = m
+	}
+
+	m.Requests++
+	m.PromptTokens += int64(promptTokens)
+	m.CompletionTokens += int64(completionTokens)
+}
+
+func (u *UsageTracker) Stats() api.UsageResponse {
+	u.mu.Lock()
+	defer u.mu.Unlock()
+
+	byModel := make([]api.ModelUsageData, 0, len(u.models))
+	for model, usage := range u.models {
+		byModel = append(byModel, api.ModelUsageData{
+			Model:            model,
+			Requests:         usage.Requests,
+			PromptTokens:     usage.PromptTokens,
+			CompletionTokens: usage.CompletionTokens,
+		})
+	}
+
+	return api.UsageResponse{
+		Start: u.start,
+		Usage: byModel,
+	}
+}

server/usage_test.go

@@ -0,0 +1,136 @@
+package server
+
+import (
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"sync"
+	"testing"
+
+	"github.com/gin-gonic/gin"
+	"github.com/ollama/ollama/api"
+)
+
+func TestUsageTrackerRecord(t *testing.T) {
+	tracker := NewUsageTracker()
+
+	tracker.Record("model-a", 10, 20)
+	tracker.Record("model-a", 5, 15)
+	tracker.Record("model-b", 100, 200)
+
+	stats := tracker.Stats()
+
+	if len(stats.Usage) != 2 {
+		t.Fatalf("expected 2 models, got %d", len(stats.Usage))
+	}
+
+	lookup := make(map[string]api.ModelUsageData)
+	for _, m := range stats.Usage {
+		lookup[m.Model] = m
+	}
+
+	a := lookup["model-a"]
+	if a.Requests != 2 {
+		t.Errorf("model-a requests: expected 2, got %d", a.Requests)
+	}
+	if a.PromptTokens != 15 {
+		t.Errorf("model-a prompt tokens: expected 15, got %d", a.PromptTokens)
+	}
+	if a.CompletionTokens != 35 {
+		t.Errorf("model-a completion tokens: expected 35, got %d", a.CompletionTokens)
+	}
+
+	b := lookup["model-b"]
+	if b.Requests != 1 {
+		t.Errorf("model-b requests: expected 1, got %d", b.Requests)
+	}
+	if b.PromptTokens != 100 {
+		t.Errorf("model-b prompt tokens: expected 100, got %d", b.PromptTokens)
+	}
+	if b.CompletionTokens != 200 {
+		t.Errorf("model-b completion tokens: expected 200, got %d", b.CompletionTokens)
+	}
+}
+
+func TestUsageTrackerConcurrent(t *testing.T) {
+	tracker := NewUsageTracker()
+
+	var wg sync.WaitGroup
+	for range 100 {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			tracker.Record("model-a", 1, 2)
+		}()
+	}
+	wg.Wait()
+
+	stats := tracker.Stats()
+	if len(stats.Usage) != 1 {
+		t.Fatalf("expected 1 model, got %d", len(stats.Usage))
+	}
+
+	m := stats.Usage[0]
+	if m.Requests != 100 {
+		t.Errorf("requests: expected 100, got %d", m.Requests)
+	}
+	if m.PromptTokens != 100 {
+		t.Errorf("prompt tokens: expected 100, got %d", m.PromptTokens)
+	}
+	if m.CompletionTokens != 200 {
+		t.Errorf("completion tokens: expected 200, got %d", m.CompletionTokens)
+	}
+}
+
+func TestUsageTrackerStart(t *testing.T) {
+	tracker := NewUsageTracker()
+
+	stats := tracker.Stats()
+	if stats.Start.IsZero() {
+		t.Error("expected non-zero start time")
+	}
+}
+
+func TestUsageHandler(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	s := &Server{
+		usage: NewUsageTracker(),
+	}
+
+	s.usage.Record("llama3", 50, 100)
+	s.usage.Record("llama3", 25, 50)
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Request = httptest.NewRequest(http.MethodGet, "/api/usage", nil)
+
+	s.UsageHandler(c)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status 200, got %d", w.Code)
+	}
+
+	var resp api.UsageResponse
+	if err := json.Unmarshal(w.Body.Bytes(), &resp); err != nil {
+		t.Fatalf("failed to unmarshal response: %v", err)
+	}
+
+	if len(resp.Usage) != 1 {
+		t.Fatalf("expected 1 model, got %d", len(resp.Usage))
+	}
+
+	m := resp.Usage[0]
+	if m.Model != "llama3" {
+		t.Errorf("expected model llama3, got %s", m.Model)
+	}
+	if m.Requests != 2 {
+		t.Errorf("expected 2 requests, got %d", m.Requests)
+	}
+	if m.PromptTokens != 75 {
+		t.Errorf("expected 75 prompt tokens, got %d", m.PromptTokens)
+	}
+	if m.CompletionTokens != 150 {
+		t.Errorf("expected 150 completion tokens, got %d", m.CompletionTokens)
+	}
+}