pi-model-info

Give your AI agent direct access to Pi's model registry — list, inspect, and check the current model — without leaving the conversation.

Install

pi install npm:pi-model-info

Requires Pi v0.28+. No other dependencies.

Tools

| Tool | What it does | |------|-------------| | model_list | List available models with fuzzy search, filter by input type, reasoning support, or provider | | model_inspect | Show full details for a specific model — cost, context window, capabilities | | model_current | Show the currently active model, thinking level, and context usage |

All tools are read-only. No model switching, no mutations.

Examples

model_list

Filter by provider to see what's available from specific sources:

List models from xiaomi and opencode-go.

Result:

xiaomi-token-plan-cn/mimo-v2-omni      MiMo-V2-Omni [reasoning, image]  ctx:262K  out:131K  in:$0.4   out:$2/1M
xiaomi-token-plan-cn/mimo-v2-pro       MiMo-V2-Pro [reasoning]          ctx:1.0M  out:131K  in:$1     out:$3/1M
xiaomi-token-plan-cn/mimo-v2.5         MiMo-V2.5 [reasoning, image]     ctx:1.0M  out:131K  in:$0.4   out:$2/1M
xiaomi-token-plan-cn/mimo-v2.5-pro     MiMo-V2.5-Pro [reasoning]        ctx:1.0M  out:131K  in:$1     out:$3/1M
opencode-go/deepseek-v4-flash          DeepSeek V4 Flash [reasoning]    ctx:1.0M  out:384K  in:$0.14  out:$0.28/1M
opencode-go/deepseek-v4-pro            DeepSeek V4 Pro [reasoning]      ctx:1.0M  out:384K  in:$1.74  out:$3.48/1M
opencode-go/glm-5                      GLM-5 [reasoning]                ctx:203K  out:33K   in:$1     out:$3.2/1M
opencode-go/glm-5.1                    GLM-5.1 [reasoning]              ctx:203K  out:33K   in:$1.4   out:$4.4/1M
opencode-go/kimi-k2.5                  Kimi K2.5 [reasoning, image]     ctx:262K  out:66K   in:$0.6   out:$3/1M
opencode-go/kimi-k2.6                  Kimi K2.6 [reasoning, image]     ctx:262K  out:66K   in:$0.95  out:$4/1M
opencode-go/mimo-v2.5                  MiMo V2.5 [reasoning, image]     ctx:1.0M  out:128K  in:$0.14  out:$0.28/1M
opencode-go/mimo-v2.5-pro              MiMo V2.5 Pro [reasoning]        ctx:1.0M  out:128K  in:$1.74  out:$3.48/1M
opencode-go/minimax-m2.5               MiniMax M2.5 [reasoning]         ctx:205K  out:66K   in:$0.3   out:$1.2/1M
opencode-go/minimax-m2.7               MiniMax M2.7 [reasoning]         ctx:205K  out:131K  in:$0.3   out:$1.2/1M
opencode-go/minimax-m3                 MiniMax M3 [reasoning, image]    ctx:512K  out:131K  in:$0.3   out:$1.2/1M
opencode-go/qwen3.6-plus               Qwen3.6 Plus [reasoning, image]  ctx:1.0M  out:66K   in:$0.5   out:$3/1M
opencode-go/qwen3.7-max                Qwen3.7 Max [reasoning]          ctx:1.0M  out:66K   in:$2.5   out:$7.5/1M
opencode-go/qwen3.7-plus               Qwen3.7 Plus [reasoning, image]  ctx:1.0M  out:66K   in:$0.4   out:$1.6/1M

model_inspect

Get the full picture of any model — API type, input modalities, pricing, compat flags:

Show details for xiaomi-token-plan-cn/mimo-v2.5.

Result:

Model:      xiaomi-token-plan-cn/mimo-v2.5
Name:       MiMo-V2.5
Provider:   xiaomi-token-plan-cn
API:        openai-completions
Reasoning:  yes
Input:      text, image
Context:    1.0M
Max output: 131K
Cost:       in:$0.4 out:$2/1M
Compat:     {"thinkingFormat":"deepseek","requiresReasoningContentOnAssistantMessages":true}

model_current

Know exactly what model is active and how much context you've burned through:

What model am I using right now?

Result:

Model:      opencode-go/deepseek-v4-flash
Name:       DeepSeek V4 Flash
Provider:   opencode-go
API:        openai-completions
Reasoning:  yes
Input:      text
Context:    1.0M
Max output: 384K
Cost:       in:$0.14 out:$0.28/1M
Thinking:   high, xhigh
Compat:     {"thinkingFormat":"deepseek","maxTokensField":"max_tokens","requiresReasoningContentOnAssistantMessages":true}
Thinking level: xhigh
Context usage: 83222 / 1000000 (8.3%)

Use cases

• Subagent orchestration — check capabilities and pricing before delegating to the right model
• Cost awareness — verify rates before running expensive prompts
• Diagnostics — confirm which model is active when behavior seems off
• Discovery — browse what's available when providers add or remove models

How it works

Pi's ModelRegistry knows every configured model — context windows, pricing, reasoning support, input types. Without this extension the agent can't ask. pi-model-info bridges that gap through three lightweight tools using Pi's extension API.

• model_list reads registry.getAvailable() with client-side fuzzy filtering (same algorithm as Pi's Ctrl+L selector)
• model_inspect does a registry.find() by provider/id, or auto-matches by id
• model_current reads the active model from context, plus optional thinking level and context usage

Development

git clone https://github.com/smilefaace/pi-model-info
cd pi-model-info
# Symlink for local testing:
pi link /path/to/pi-model-info

Publish:

# Bump version in package.json
npm publish

Changelog

1.0.2

• Richer README with real usage examples and screenshots

1.0.1

• Improved error messages for empty results and ambiguous model IDs
• Better truncation logic when model list exceeds display limit

1.0.0

• Initial release: model_list, model_inspect, model_current

License

MIT