Baseten Provider Setup for Inkling and Model APIs

Baseten's Model APIs give hosted, OpenAI-compatible entry points to frontier models. Because the official external plugin relies on authenticated discovery, OpenClaw picks up the full set of models enabled on your Baseten account. When no credentials are present, the offline fallback ships with every Model API that existed at the time this OpenClaw build was released.

Install plugin

openclaw plugins install @openclaw/baseten-provider
openclaw gateway restart

Getting started

Create a Baseten account and API key

There is no monthly platform charge on Baseten's Basic plan; Model API usage is billed per call. Head to Baseten API key settings to generate a key, and consult the pricing page for up-to-date rates.

Run onboarding

openclaw onboard --auth-choice baseten-api-key

openclaw onboard --non-interactive --accept-risk --skip-health \
  --auth-choice baseten-api-key \
  --baseten-api-key "$BASETEN_API_KEY"

export BASETEN_API_KEY=...

Verify the live catalog

openclaw models list --provider baseten

When authentication works, the plugin requests GET /v1/models and enumerates every model the account returns. Without auth, it remains offline and relies on the bundled fallback list.

Inkling

The default model is Inkling from Thinking Machines Lab. Within OpenClaw it handles text and image inputs, tool calling, structured tool schemas, adjustable reasoning effort, a context window of 1.048M tokens, and output of up to 32k tokens:

{
  agents: {
    defaults: {
      model: { primary: "baseten/thinkingmachines/inkling" },
    },
  },
}

To switch the active session, use /model baseten/thinkingmachines/inkling -s.

Bundled fallback catalog

The live catalog seen after authentication is the source of truth. Until discovery succeeds, these rows keep configuration and model choice workable:

Reasoning and tool calling are supported across all bundled models. OpenClaw translates its thinking levels into native reasoning_effort on models that offer it. Baseten's opt-in GLM, Kimi, and Nemotron models ship with thinking disabled by default; most give a simple off/on toggle, while GLM 5.2 adds off, high, and max. Those choices are relayed via Baseten's chat_template_args.enable_thinking control, and for GLM 5.2 the validated top-level reasoning_effort parameter is used as well.

Note

Model APIs on Baseten can appear, disappear, or change without any connection to OpenClaw releases. The plugin pulls fresh model ids, context caps, output caps, and input, cached-input, and output pricing from the authenticated API, all while keeping model-specific OpenClaw transport policy intact.

Manual config

For most cases the API key alone is enough. To force the provider explicitly:

{
  env: { vars: { BASETEN_API_KEY: "..." } },
  agents: {
    defaults: {
      model: { primary: "baseten/thinkingmachines/inkling" },
    },
  },
  models: {
    mode: "merge",
    providers: {
      baseten: {
        baseUrl: "https://inference.baseten.co/v1",
        apiKey: "${BASETEN_API_KEY}",
        api: "openai-completions",
        models: [
          {
            id: "thinkingmachines/inkling",
            name: "Inkling",
            reasoning: true,
            input: ["text", "image"],
            contextWindow: 1048000,
            maxTokens: 32000,
          },
        ],
      },
    },
  },
}

Note

When the Gateway runs as a daemon (launchd, systemd, Docker), confirm that BASETEN_API_KEY is visible to that process. A key set only in an interactive shell won't be seen by a managed service that's already running.

Related

• Model providers, Picking providers, model refs, and failover behavior.

• Thinking modes, Choose OpenClaw reasoning effort levels.

• Models CLI, Show, inspect, and pick discovered models.

• Models FAQ, Auth profiles and model-selection troubleshooting.