Skip to main content

LLM providers

EnClaw's agents and chat all run on LLMs. Without at least one provider configured, the workspace is empty plumbing — no model means no completions, no embeddings, no streaming responses.

This is the smallest step in the setup checklist — usually 5 minutes.

Pick a starting provider​

Any one of these will get you running. You can add more later — agents and squads can route to specific providers, but the platform doesn't require multiples.

ProviderBest forDetailed guide
AnthropicClaude models directdocs/integrations/llmproviders/anthropic.md (in the repository)
Azure OpenAIGPT-4 family via your existing Azure tenantdocs/integrations/llmproviders/azure-openai.md (in the repository)
OpenAIGPT-4 family direct from OpenAIdocs/integrations/llmproviders/openai.md (in the repository)
GoogleGemini familydocs/integrations/llmproviders/google.md (in the repository)
AWS BedrockClaude / Llama via Bedrockdocs/integrations/llmproviders/bedrock.md (in the repository)
Azure AI FoundryMulti-model via Azuredocs/integrations/llmproviders/azure-ai.md (in the repository)
DeepSeekOpen-source alternativedocs/integrations/llmproviders/deepseek.md (in the repository)
OllamaLocal self-hosted modelsdocs/integrations/llmproviders/ollama.md (in the repository)
LMStudioLocal self-hosted modelsdocs/integrations/llmproviders/lmstudio.md (in the repository)
Custom OpenAI-compatibleAnything that speaks the OpenAI Chat APIdocs/integrations/llmproviders/custom.md (in the repository)

If your org is "all-in on Azure", Azure OpenAI is the path of least resistance — billing rolls up to your existing Azure subscription, no third-party signup. If you want Claude specifically and don't mind a second vendor, Anthropic is the simplest setup (single API key).

Full provider matrix + pros/cons: docs/integrations/llmproviders/README.md (in the repository).

Quick setup (Anthropic example)​

  1. Get an API key from https://console.anthropic.com → API Keys → Create Key.
  2. In EnClaw: /integrations → LLM Providers → + Add provider.
  3. Pick Anthropic from the dropdown.
  4. Paste the API key.
  5. Optionally cap rate limits and monthly budget so a runaway agent can't blow through your quota.
  6. Tick Is active and Is default (the default provider is used when an agent doesn't specify one).
  7. Save → click Test → expect a green "connection verified".

Pool vs tenant-scoped​

If you're on the managed SaaS, your operator may have configured a shared LLM pool — meaning you don't need to bring your own key. In that case the /integrations → LLM Providers page will show pool entries with tenant=platform and you can use them without adding anything. Talk to your operator if you're unsure whether your tenant has pool access.

You can still add tenant-scoped providers on top of pool ones — useful if you want a higher-tier Claude key just for your power users.

Per-agent / per-team override​

The default provider above kicks in for every new agent. Individual agents can override:

  • Agent → Edit → Model picker — choose a specific provider + model.
  • Team → Default model — sets a per-team default that overrides tenant-default for agents in that team.

Setting tenant-default first then overriding for specific teams / agents is the cleanest mental model.

What goes wrong​

  • "No models found" in the model picker → the provider didn't return a usable model list. Click Test on the provider row — the error message usually says auth failed, region mismatch, or quota exhausted.
  • Agent returns "Provider not configured" mid-conversation → the provider got toggled inactive or its credentials rotated. Re-test in /integrations and reactivate.
  • Streaming responses cut off after 5 min → check the agent's timeout settings; some providers throttle long-running streams. Workaround is to disable streaming for that agent.

Next​