cmd: add usage cmd to chat to see token consumption

Adding a `/usage` command to interactive cli chat sessions that displays the tokens used in the current sessions. This can be used alongside the models context window to understand when a context shift is going to happen.

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, error) {
+func chat(cmd *cobra.Command, opts runOptions) (*api.Message, *api.Metrics, error) {
 	client, err := api.ClientFromEnvironment()
 	if err != nil {
-		return nil, err
+		return nil, nil, err
 	}
 
 	p := progress.NewProgress(os.Stderr)
@@ -1515,7 +1515,7 @@ func chat(cmd *cobra.Command, opts runOptions) (*api.Message, error) {
 
 	if err := client.Chat(cancelCtx, req, fn); err != nil {
 		if errors.Is(err, context.Canceled) {
-			return nil, nil
+			return nil, 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, error) {
 			p.StopAndClear()
 			fmt.Println("An error occurred while processing your message. Please try again.")
 			fmt.Println()
-			return nil, nil
+			return nil, nil, nil
 		}
-		return nil, err
+		return nil, nil, err
 	}
 
 	if len(opts.Messages) > 0 {
@@ -1535,14 +1535,14 @@ func chat(cmd *cobra.Command, opts runOptions) (*api.Message, error) {
 
 	verbose, err := cmd.Flags().GetBool("verbose")
 	if err != nil {
-		return nil, err
+		return nil, nil, err
 	}
 
 	if verbose {
 		latest.Summary()
 	}
 
-	return &api.Message{Role: role, Thinking: thinkingContent.String(), Content: fullResponse.String()}, nil
+	return &api.Message{Role: role, Thinking: thinkingContent.String(), Content: fullResponse.String()}, &latest.Metrics, 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/selector.go

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

cmd/interactive.go

@@ -30,6 +30,9 @@ 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")
@@ -37,6 +40,7 @@ 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")
@@ -445,6 +449,9 @@ 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 {
@@ -499,7 +506,7 @@ func generateInteractive(cmd *cobra.Command, opts runOptions) error {
 
 			opts.Messages = append(opts.Messages, newMessage)
 
-			assistant, err := chat(cmd, opts)
+			assistant, metrics, err := chat(cmd, opts)
 			if err != nil {
 				if strings.Contains(err.Error(), "does not support thinking") ||
 					strings.Contains(err.Error(), "invalid think value") {
@@ -509,6 +516,10 @@ 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/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 | bash
+curl -fsSL https://opencode.ai/install.sh | 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>

model/parsers/qwen3coder.go

@@ -19,9 +19,8 @@ import (
 type qwenParserState int
 
 const (
-	toolOpenTag       = "<tool_call>"
-	toolCloseTag      = "</tool_call>"
-	functionOpenStart = "<function=" // qwen3-coder sometimes omits <tool_call> but starts with this
+	toolOpenTag  = "<tool_call>"
+	toolCloseTag = "</tool_call>"
 )
 
 const (
@@ -139,26 +138,11 @@ func eat(p *Qwen3CoderParser) ([]qwenEvent, bool) {
 			p.acc.WriteString(after)
 			p.state = qwenParserState_CollectingToolContent
 			return events, true
-		} else if idx := strings.Index(p.acc.String(), functionOpenStart); idx != -1 {
-			// qwen3-coder sometimes omits <tool_call> but starts with <function=
-			// we treat this as the start of a tool call, keeping the <function= prefix
-			// since it's part of the raw tool call content
-			before := p.acc.String()[:idx]
-			before = strings.TrimRightFunc(before, unicode.IsSpace)
-			if len(before) > 0 {
-				events = append(events, qwenEventContent{content: before})
-			}
-			after := p.acc.String()[idx:]
-			p.acc.Reset()
-			p.acc.WriteString(after)
-			p.state = qwenParserState_CollectingToolContent
-			return events, true
-		} else if toolOverlap, funcOverlap := overlap(p.acc.String(), toolOpenTag), overlap(p.acc.String(), functionOpenStart); toolOverlap > 0 || funcOverlap > 0 {
+		} else if overlap := overlap(p.acc.String(), toolOpenTag); overlap > 0 {
 			// we found a partial tool open tag, so we can emit the unambiguous part,
 			// which is the (trailing-whitespace trimmed) content before the partial
 			// tool open tag
-			maxOverlap := max(toolOverlap, funcOverlap)
-			beforePartialTag := p.acc.String()[:len(p.acc.String())-maxOverlap]
+			beforePartialTag := p.acc.String()[:len(p.acc.String())-overlap]
 			trailingWhitespaceLen := trailingWhitespaceLen(beforePartialTag)
 			ambiguousStart := len(beforePartialTag) - trailingWhitespaceLen
 			unambiguous := p.acc.String()[:ambiguousStart]

model/parsers/qwen3coder_test.go

@@ -343,23 +343,20 @@ func TestQwenParserStreaming(t *testing.T) {
 				},
 			},
 		},
-		// qwen3-coder:30b occasionally leaves off opening <tool_call> tags, but we
-		// want to parse it anyway
-		{
-			desc: "missing <tool_call> opening tag still parses",
-			steps: []step{
-				{
-					input: "before tool call<function=get_current_temperature>some tool content here</function></tool_call>",
-					wantEvents: []qwenEvent{
-						qwenEventContent{content: "before tool call"},
-						qwenEventRawToolCall{raw: "<function=get_current_temperature>some tool content here</function>"},
-					},
-				},
-			},
-		},
+	}
+
+	anyOnlies := false
+	for _, tc := range cases {
+		if tc.only {
+			anyOnlies = true
+		}
 	}
 
 	for _, tc := range cases {
+		if anyOnlies && !tc.only {
+			continue
+		}
+
 		t.Run(tc.desc, func(t *testing.T) {
 			parser := Qwen3CoderParser{}