Using Ollama (Local LLMs) with OpenClaw

Prerequisites

• ✓Ollama installed on your machine or server — install from https://ollama.com/download.
• ✓Ollama running and reachable on the default port `http://127.0.0.1:11434` or another host/port you control.
• ✓The OpenClaw CLI installed and working so you can run `openclaw onboard`, `openclaw config`, and `openclaw models` commands.
• ✓At least one Ollama model pulled locally (for example `gemma4`, `gpt-oss:20b`, or `llama3.3`) or access to Ollama cloud models via `ollama signin`.

Steps

1. 1

   Install and run Ollama

   Start by installing Ollama and making sure the server is running, otherwise OpenClaw has nothing to talk to. You install from the Ollama site, then run the Ollama service so the native `/api/chat` and `/api/tags` endpoints are available on port 11434.

   bashCopy

   ollama serve

2. 2

   Pull local models and optionally sign in for cloud models

   Pull at least one local model so OpenClaw has something to auto-discover, and sign in if you also want Ollama cloud models. 5:cloud` do not require a local pull, but local ones do.

   bashCopy

   ollama pull gemma4
   # or
   ollama pull gpt-oss:20b
   # or
   ollama pull llama3.3
   
   ollama signin

3. 3

   Run OpenClaw onboarding with Ollama

   Use the OpenClaw onboarding wizard to wire up Ollama without hand-editing config. This step lets you choose between Local or Cloud + Local, sets the base URL, and can auto-pull a selected model if it is missing.

   bashCopy

   openclaw onboard

4. 4

   Automate onboarding in non-interactive environments

   For CI or scripted setups, run onboarding in non-interactive mode so you can bake Ollama into your deployment. You can also pass a custom base URL and model ID when your Ollama server runs on another host or you want a specific default model.

   bashCopy

   openclaw onboard --non-interactive \
     --auth-choice ollama \
     --accept-risk
   
   openclaw onboard --non-interactive \
     --auth-choice ollama \
     --custom-base-url "http://ollama-host:11434" \
     --custom-model-id "qwen3.5:27b" \
     --accept-risk

5. 5

   Enable Ollama via environment variable for implicit discovery

   If you prefer minimal config, enable Ollama by setting `OLLAMA_API_KEY` and let OpenClaw auto-discover models from your local instance. ollama` if you want this implicit discovery behavior.

   bashCopy

   export OLLAMA_API_KEY="ollama-local"

6. 6

   Configure Ollama explicitly when running on a remote host

   When Ollama runs on another host or you want full control over model definitions, configure the provider explicitly. Use the native base URL without `/v1` so you keep native tool calling and streaming behavior.

   jsonCopy

   {
     models: {
       providers: {
         ollama: {
           baseUrl: "http://ollama-host:11434",
           apiKey: "ollama-local",
           api: "ollama",
           models: [
             {
               id: "gpt-oss:20b",
               name: "GPT-OSS 20B",
               reasoning: false,
               input: ["text"],
               cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
               contextWindow: 8192,
               maxTokens: 8192 * 10
             }
           ]
         }
       }
     }
   }

7. 7

   Set a custom base URL without defining models

   If you only need to point OpenClaw at a different Ollama host and still rely on native behavior, use an explicit provider with `baseUrl` and `api`. This disables auto-discovery, so plan to define models manually if you need a catalog.

   jsonCopy

   {
     models: {
       providers: {
         ollama: {
           apiKey: "ollama-local",
           baseUrl: "http://ollama-host:11434", // No /v1 - use native Ollama API URL
           api: "ollama", // Set explicitly to guarantee native tool-calling behavior
         },
       },
     },
   }

8. 8

   Choose default and fallback Ollama models for your agents

   Once Ollama is wired up, configure which model your agents use by default and what they fall back to. This keeps your agents resilient if a model is missing or fails.

   jsonCopy

   {
     agents: {
       defaults: {
         model: {
           primary: "ollama/gpt-oss:20b",
           fallbacks: ["ollama/llama3.3", "ollama/qwen2.5-coder:32b"],
         },
       },
     },
   }

9. 9

   Set the default Ollama model via CLI or config

   Use the CLI to inspect available models and switch the active one, or hard-code the default in your config. This is useful when you want a specific local model like `gemma4` to be the primary for all agents.

   bashCopy

   openclaw models list
   openclaw models set ollama/gemma4
   
   {
     agents: {
       defaults: {
         model: { primary: "ollama/gemma4" },
       },
     },
   }

10. 10

    Enable Ollama Web Search as the web search provider

    If you want your agents to call Ollama’s web search, configure it as the `web_search` provider. This uses your existing Ollama host, requires `ollama signin`, and stays key-free.

    jsonCopy

    {
      tools: {
        web: {
          search: {
            provider: "ollama",
          },
        },
      },
    }

11. 11

    Use reasoning models with Ollama

    For reasoning-style models, pull an Ollama model whose name matches the reasoning heuristic so OpenClaw treats it as reasoning-capable. This lets you experiment with models like `deepseek-r1` while keeping config simple.

    bashCopy

    ollama pull deepseek-r1:32b

12. 12

    Configure legacy OpenAI-compatible mode when required by a proxy

    If you sit behind a proxy that only speaks OpenAI format, switch Ollama to OpenAI-compatible mode in OpenClaw. Be aware that tool calling is not reliable here and you may need to disable streaming or `injectNumCtxForOpenAICompat` depending on your upstream.

    jsonCopy

    {
      models: {
        providers: {
          ollama: {
            baseUrl: "http://ollama-host:11434/v1",
            api: "openai-completions",
            injectNumCtxForOpenAICompat: true, // default: true
            apiKey: "ollama-local",
            models: [...]
          }
        }
      }
    }
    
    {
      models: {
        providers: {
          ollama: {
            baseUrl: "http://ollama-host:11434/v1",
            api: "openai-completions",
            injectNumCtxForOpenAICompat: false,
            apiKey: "ollama-local",
            models: [...]
          }
        }
      }
    }

Configuration

Troubleshooting

ollama serve

curl http://localhost:11434/api/tags

ollama list  # See what's installed
ollama pull gemma4
ollama pull gpt-oss:20b
ollama pull llama3.3     # Or another model

# Check if Ollama is running
ps aux | grep ollama

# Or restart Ollama
ollama serve

Frequently asked questions