Topchester
Changelog Docs

Models and providers

Model references use provider/model-id. The first slash separates the provider, so nested model IDs such as openrouter/google/gemini-3.1-flash-lite work as-is.

For a temporary selection, no model config is needed:

OPENROUTER_API_KEY=... topchester -m openrouter/google/gemini-3.1-flash-lite

OpenRouter and Codex are built-in providers. Topchester can materialize their provider defaults in memory when a CLI or slash-command reference uses them. Custom provider IDs must be defined under providers in JSONC.

Topchester has three user-facing model slots:

  • default runs the main agent and fills in unspecified model work.
  • fast runs quick checks and lightweight agent calls.
  • kb.summarize runs knowledge-base summarization.

If fast or kb.summarize is omitted, Topchester uses default.

One OpenRouter model

{
  "models": {
    "default": "openrouter/google/gemini-3.1-flash-lite",
  },
}

Separate fast and KB models

{
  "models": {
    "default": "openrouter/anthropic/claude-sonnet-4.5",
    "fast": "openrouter/google/gemini-3.1-flash-lite",
    "kb.summarize": "openrouter/google/gemini-3.1-pro",
  },
}

Custom provider

Use this shape for OpenAI-compatible endpoints such as OpenRouter, LiteLLM, vLLM, LM Studio, Ollama, or local OpenAI-compatible proxies:

{
  "models": {
    "default": "anthropic/claude-sonnet-4.5",
  },
  "providers": {
    "default": "openrouter",
    "openrouter": {
      "type": "openai-compatible",
      "baseURL": "https://openrouter.ai/api/v1",
      "apiKeyEnv": "OPENROUTER_API_KEY",
      "supportsStructuredOutputs": true,
    },
  },
}

Prefer apiKeyEnv over apiKey so secrets stay out of config files.

Interactive selection

/model provider/model-id selects that exact model for the current session, even when it is not in models.choices. Bare /model and /models open the saved choices picker. /model all [search] browses the OpenRouter catalog and can add a choice. /effort and /reasoning set a provider-level effort override for the current session. These controls never write the effective merged config back to disk.

The root -m, --model option and topchester run -m, --model use the same reference rules. Session model and effort selections survive --resume, /restore, and /fork; an explicit CLI model wins over the restored model. /new starts from the loaded JSONC defaults. For durable defaults, edit models.default or the provider's reasoningEffort in the intended JSONC file.

For a local OpenAI-compatible proxy, including VibeProxy, a selected profile can contain:

{
  "models": {
    "default": { "name": "gpt-5.5(low)", "provider": "openai" },
    "choices": ["openai/gpt-5.5(low)", "openai/gpt-5.5(high)"],
  },
  "providers": {
    "default": "openai",
    "openai": {
      "type": "openai-compatible",
      "baseURL": "http://127.0.0.1:8317/v1",
      "apiKey": "dummy-not-used",
    },
  },
}

Start it with topchester --config ./vibeproxy.jsonc. A model-id effort suffix remains part of the model name; a Topchester /effort override is also shown separately in the status line, and proxy-specific precedence remains the proxy's responsibility.