#Not getting a response after installing

1 messages · Page 1 of 1 (latest)

atomic sand
#

Hi, I installed openclaw with the one liner on the main website, and went through the configs, inputted my api key (kimi 2.5). I’ve used the UI and the TUI, but am not able to get a response from the ai. It’s just blank. Any ideas?

mild vergeBOT
#

If both the Dashboard UI and the TUI show “nothing” after you send a message, it usually means the gateway is running but the model call is failing (auth/provider mismatch, 401, wrong model ref), and the error is only visible in logs.

Can you run these and paste the output (redact keys/tokens)?

openclaw --version
openclaw gateway status
openclaw models status --probe
openclaw doctor

Then, in another terminal:

openclaw logs --follow
#

Send one message in the UI/TUI and paste the few log lines around the failure (that’s the smoking gun).

Kimi-specific gotcha (very common)

“Kimi 2.5” can mean two different providers in OpenClaw, and the keys are not interchangeable:

  • Moonshot (Kimi API) → model refs look like moonshot/... (e.g. moonshot/kimi-k2.5)
  • Kimi Coding → model refs look like kimi-coding/... (e.g. kimi-coding/k2p5)

Docs: https://docs.openclaw.ai/providers/moonshot

So in openclaw models status --probe, check:

  1. which provider it resolved as the default, and
  2. whether the probe is returning 401/403 (wrong key/provider) vs something else.

If you paste the models status --probe output + the log snippet, I can tell you exactly which side is misconfigured (wrong provider, wrong baseUrl like .cn vs .ai, wrong model ref, etc.).

#

Your gateway looks healthy (RPC probe: ok), and OpenClaw is actually resolving a default model (kimi-coding/k2p5). So the “blank reply” is almost certainly the model request failing (and the error will show up in logs).

Two things jump out:

  1. Provider/key mismatch is very likely
  • You’re configured for Kimi Coding: kimi-coding/k2p5
  • The stored key prefix shows as nvapi-… which doesn’t look like a Moonshot/Kimi key. It looks more like an NVIDIA API key.
  • If you meant Moonshot (Kimi API / open platform), you want moonshot/kimi-k2.5 instead (different provider, different key, different endpoint).

Docs (and the important “Moonshot vs Kimi Coding are separate” note):
https://docs.openclaw.ai/providers/moonshot

  1. We need the actual error line from logs
    Please run these and paste outputs (redact secrets):
# 
openclaw doctor
openclaw logs --limit 200

Then do a direct “one-shot” agent call (bypasses the UI/TUI, super helpful to isolate):

openclaw agent --message "Reply with only the word: pong"

If you can also answer this: where did you get the API key from (Moonshot Open Platform vs Kimi Coding vs OpenRouter/Together/etc.)?
Once I know that + the log snippet, I can tell you exactly what to change (usually it’s just switching to moonshot/kimi-k2.5 or re-onboarding with the correct Kimi Coding key).

#

That explains it: an nvapi-… key is NVIDIA API Catalog, but your OpenClaw config is currently using the kimi-coding provider (kimi-coding/k2p5).

Those are different systems, so OpenClaw will send the request to Kimi Coding’s endpoint with an NVIDIA key → the model call fails (often looks like “blank” in UI unless you check logs).

You’ve got two options:

Option A (recommended): use the “real” Kimi provider for OpenClaw

If you actually want Kimi/K2.5 directly, use Moonshot or Kimi Coding keys (not NVIDIA):

Option B: keep the NVIDIA key (manual custom provider)

If you specifically want to use NVIDIA’s hosted Kimi, set up a custom OpenAI-compatible provider and point it at NVIDIA’s base URL + model id (whatever NVIDIA calls it in their docs/model card).

#

In ~/.openclaw/openclaw.json, add something like:

{
  env: { NVIDIA_API_KEY: "nvapi-..." },

  agents: {
    defaults: {
      model: { primary: "nvidia/<NVIDIA_MODEL_ID>" },
    },
  },

  models: {
    mode: "merge",
    providers: {
      nvidia: {
# 
        baseUrl: "https://integrate.api.nvidia.com/v1", // confirm from NVIDIA docs
        apiKey: "${NVIDIA_API_KEY}",
        api: "openai-completions",
        models: [{ id: "<NVIDIA_MODEL_ID>", name: "Kimi (NVIDIA)" }],
      },
    },
  },
}

Key point: <NVIDIA_MODEL_ID> must match NVIDIA’s model name (grab it from their model page/docs).

Next: confirm the exact failure in logs

Run: