Models

Clawpy is designed around bring-your-own-key model routing. You can assign different providers and runtime profiles to different agents, lanes, and risk levels without locking the system to one fixed model list.

Supported Provider Families

The exact model catalogue changes over time, but Clawpy is intended to work with major hosted and local provider families through its routing and vault layers:

Provider familyTypical use
OpenAI-compatible APIsgeneral reasoning, coding, tool use, structured outputs
Anthropic-compatible APIsplanning, code review, writing, careful reasoning
Google Gemini-compatible APIsresearch, multimodal, fast triage, broad context
DeepSeek, Qwen, Moonshot, MiniMax, and similar providerscost-aware reasoning, coding, fallback, regional/provider diversity
Local or self-hosted runtimesprivacy-sensitive work, cheap triage, offline/local experimentation

How Model Selection Works

Each agent or workflow lane can use a different model profile. That lets you optimise for capability, cost, latency, privacy, and risk:

  • Fast triage models for routing, classification, and low-risk summaries
  • Reasoning models for architecture, review, planning, and disputed decisions
  • Coding models for Clawpito review, scoped fixes, tests, and release-readiness work
  • Local models for privacy-sensitive or low-cost background support

Select and tune models in the dashboard vault and routing surfaces. Where a workflow is high risk, Guardian policy and human review should matter more than raw model capability.

Bring Your Own Key

Clawpy uses a BYOK model. You add provider credentials to the Vault, and costs are billed by the provider you configure. Clawpy should not require one blessed model stack to be useful.

Routing And Budgets

Under the hood, Clawpy normalises provider access through routing services so agents can share common budget, telemetry, retry, and fallback behavior. The intent is to make model choice a policy decision rather than a hard-coded implementation detail.