How to Use Kimi with OpenClaw

Short answer

For the Moonshot API route, OpenClaw’s Moonshot provider plugin connects to Kimi. The setup guide used here is based on OpenClaw 2026.7.1 and selects moonshot/kimi-k3.1-pro; check your installed model catalog before choosing a model. Kimi’s guide says its hosted Kimi K2.5 offering was sunset on August 31, 2026, so K2.5 is not the model used in this walkthrough. That statement concerns Kimi’s hosted platform; it does not establish availability through every independent gateway or local deployment. (Kimi’s OpenClaw guide, OpenClaw Moonshot guide)

Moonshot API and Kimi Coding are separate providers. Use a Moonshot API key with a moonshot/... model reference; don’t substitute Kimi Coding credentials or model IDs. (OpenClaw Moonshot guide)

Prerequisites

• OpenClaw 2026.7.1, the version named by Kimi’s setup guide. If you use another release, check its provider and onboarding documentation before applying these commands.
• A Kimi API key for the China-platform route shown below. Use the option matching your account’s region; don’t enter a key from a different region into this route.
• A shell on the Gateway host. The credential workflow below is specifically for running the Gateway in the foreground from that same shell; it does not configure a separately installed background service.

Connect Kimi to OpenClaw

1. Install the Moonshot provider plugin and restart the Gateway:

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

   The plugin reference identifies @openclaw/moonshot-provider as the package. Kimi’s guide gives the install-and-restart sequence. (Plugin reference, Kimi’s setup guide)

2. Supply the key to onboarding and the foreground Gateway without putting its value in a command. In the shell you will use to launch the Gateway, read the key without echoing it and export it for child processes:

   read -r -s -p 'Moonshot API key: ' MOONSHOT_API_KEY
   export MOONSHOT_API_KEY
   printf '\n'
   

   Then run onboarding in that shell:

   openclaw onboard --auth-choice moonshot-api-key-cn --secret-input-mode ref
   

   When prompted for credential handling, choose an environment-backed reference for MOONSHOT_API_KEY, not plaintext key storage. OpenClaw documents reference mode as writing environment-backed references; Kimi’s API quickstart names MOONSHOT_API_KEY. Prompt wording can vary by release, so if onboarding does not offer that reference choice, stop and check the documentation for your installed version rather than entering the key as plaintext. (Onboarding reference, Kimi API quickstart)

   Keep this Gateway in the foreground. If onboarding offers to install or run a background service, decline that option for this workflow. Start the Gateway from the same shell that holds the exported variable:

   openclaw gateway run
   

   A Gateway launched from a different shell or as a service will not automatically inherit this shell’s environment. This procedure does not cover how to inject the variable into a particular service manager. (Run the Gateway, Environment variables)

3. In another terminal, inspect the catalog and choose an available model:

   openclaw models list --provider moonshot
   openclaw models set moonshot/kimi-k3.1-pro
   

   Use the Kimi guide’s model ID only if it appears in your installed catalog. If it does not, check the plugin and release documentation for your installation rather than guessing an ID. (Kimi’s setup guide, OpenClaw models CLI)

Protect the API key

Run the secrets audit after onboarding:

openclaw secrets audit --check

A clean check reports no findings in the surfaces it scans; it does not prove that no other copy of the key exists. The documented exit codes distinguish findings (1) from unresolved references (2). OpenClaw warns that plaintext credentials in files an agent can inspect may remain agent-readable. (Secrets CLI, Secrets management)

If you replace the key, stop the foreground Gateway with Ctrl+C, read and export the replacement in the shell you will use to launch it, then run openclaw gateway run again. This starts a new Gateway process with the updated environment. If a key is exposed, revoke it in the Kimi platform and create a replacement before relaunching; Kimi documents revocation for inference API keys. (Kimi key-revocation documentation)

Verify the connection

Use an agent ID that already exists in your OpenClaw configuration. main is an example; replace it if your configured agent has another ID. Send a test turn through the Gateway:

openclaw agent --agent main --message "Reply with exactly: Kimi is connected."

A completed response and a successful command exit show that the turn completed; the requested words alone do not prove which model answered. The --agent main selector targets that agent’s main session. Open the Control UI, select the same agent and its main conversation, and enter:

/model status

That session-level status lets you inspect the model selected for the conversation. Separately, check the configuration for the same agent:

openclaw models status --agent main --json

This reports that agent’s configured default and fallbacks; it is not upstream Moonshot telemetry, and a fallback may have served a completed turn. (Agent CLI, Models CLI, Model selection)

Troubleshoot common problems

• Model not found: Run openclaw models list --provider moonshot and use an exact listed ID.
• Authentication failure: Check that the key matches the China-platform onboarding choice and that MOONSHOT_API_KEY is available in the environment inherited by the foreground Gateway. Moonshot and Kimi Coding credentials are not interchangeable.
• Plugin not recognized: Confirm installation completed, then restart the Gateway.
• Secrets audit reports findings: Review its reported locations and resolve plaintext findings or unresolved references; rerun openclaw secrets audit --check. A clean audit covers only the surfaces it scans.
• A reply arrives but the route is uncertain: Check /model status in the same agent’s main conversation and compare it with openclaw models status --agent <agent-id> --json, including configured fallbacks. A completed turn by itself does not identify the upstream model.

Bottom line: Install the Moonshot plugin, use reference mode for the API key, launch the foreground Gateway from the environment that provides that key, and verify both turn completion and the selected model in the same agent session. This walkthrough uses Kimi K3.1 Pro when it appears in your catalog; it does not use Kimi K2.5.

Sources

在 OpenClaw 中连接 Kimi — Kimi API 开放平台

Moonshot AI — OpenClaw

Moonshot plugin reference — OpenClaw

CLI setup reference — OpenClaw

Quickstart — Kimi API Platform

Secrets management — OpenClaw

Secrets CLI — OpenClaw

Run the Gateway — OpenClaw

Environment variables — OpenClaw

Revoke inference API key — Kimi API Platform

Agent — OpenClaw

Models — OpenClaw

Models concepts — OpenClaw