Model Management

Viewing Status

/model

Shows: current model, provider, transport mode, context usage, all agent assignments, fallback chain, hailMary config.

Discovering Models

/model list              → list all providers with model counts
/model list ollama       → list models available on ollama
/model list openrouter   → list models available on openrouter

Works on all platforms (text output).

Switching Models

Interactive (Telegram)

/model change

3-step picker:

1. Pick agent slot (professor, dreamy, browsie, coding)
2. Pick provider (shows model count, current marked ✅)
3. Pick model

Quick switch (all platforms)

/model quick kimi-k2.5:cloud

Switches professor to the specified model on the current provider.

Undo / Reset

Command	Effect
/model restore	Swap transport.json ↔ transport.json.old (undo last change)
/model default	Copy transport.default.json → transport.json (factory reset)
/model health reset	Clear all model health penalties

Emergency Mode

/model emergency

Activates the paid hailMary model (configured in transport.json). Never auto-triggered — manual operator decision only. Clears on /model restore, /model default, or /reset.

Fallback Behavior

When a model fails, the bridge automatically tries the next fallback:

professor (primary)
  → fallback 1 (e.g. ollama/kimi-k2.5)
  → fallback 2 (e.g. openrouter/gemini-2.5-flash)
  → fallback N ...
  → all exhausted → error

The health system tracks errors per model:

• Rate limit (429) — model skipped, respects Retry-After header
• Auth error (401/403) — model permanently skipped until manual reset
• Transient (500, timeout) — progressive backoff, auto-recovers over time

Use /model health reset to clear all penalties and make all models available again.

Adding Models via Config Files

Two files control model availability:

~/.abtars/config/models.json — Model catalog

Defines all available models and their properties:

{
  "models": {
    "ollama": [
      { "id": "kimi-k2.5:cloud", "contextWindow": 131072 },
      { "id": "qwen3:32b", "contextWindow": 32768 }
    ],
    "openrouter": [
      { "id": "deepseek/deepseek-v4-flash", "contextWindow": 65536 },
      { "id": "google/gemini-2.5-flash", "contextWindow": 1048576 }
    ]
  }
}

Add a model here to make it available in /model list and the interactive picker.

~/.abtars/config/transport.json — Active assignments

Defines which model each agent slot uses:

{
  "agents": {
    "professor": {
      "provider": "ollama",
      "model": "kimi-k2.5:cloud",
      "fallbacks": ["openrouter/deepseek/deepseek-v4-flash"]
    },
    "dreamy": {
      "provider": "openrouter",
      "model": "deepseek/deepseek-v4-flash"
    }
  },
  "providers": {
    "ollama": { "baseUrl": "http://localhost:11434" },
    "openrouter": { "baseUrl": "https://openrouter.ai/api/v1", "apiKeyEnv": "OPENROUTER_API_KEY" }
  }
}

Adding a new provider

1. Add the provider to transport.json → providers:

   "anthropic": { "baseUrl": "https://api.anthropic.com/v1", "apiKeyEnv": "ANTHROPIC_API_KEY" }

2. Drop the API key: echo "sk-ant-..." > ~/.abtars/secret/ANTHROPIC_API_KEY
3. Add models to models.json:

   "anthropic": [{ "id": "claude-sonnet-4.6", "contextWindow": 200000 }]

4. Restart: /restart or abtars restart

Adding a new model to an existing provider

1. Add to models.json under the provider
2. Switch via /model quick <model-id> or /model change

No restart needed for switching — only for adding new providers (needs API key loaded).

Two methods compared

Method	When to use
/model change or /model quick	Switch between already-configured models (instant, no restart)
Edit transport.json + models.json	Add new providers, new models, change fallback chains, modify context windows