Nvidia API | Friends of the Crustacean 🦞🤝 | Page 1

reef kindle Apr 27, 2026, 12:35 PM

#

I've put the nvidia api, but don't know what model to choose from the list in openclaw models configuration

pallid rockBOT Apr 27, 2026, 12:35 PM

#

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

velvet spokeBOT Apr 27, 2026, 12:35 PM

#

I’d start with the NVIDIA Nemotron 3 Super 120B option — that’s the one the OpenClaw NVIDIA docs use as the default example:

openclaw models set nvidia/nvidia/nemotron-3-super-120b-a12b

If you’re running from source, use:

pnpm openclaw models set nvidia/nvidia/nemotron-3-super-120b-a12b

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

If that model doesn’t show up in your picker, paste your openclaw --version and the NVIDIA entries from openclaw models list. For broader “which model feels best right now?” advice, #models is the better place since model quality changes fast.

reef kindle Apr 27, 2026, 4:17 PM

#

is this running local or api?

velvet spokeBOT Apr 27, 2026, 4:18 PM

#

API — NVIDIA in OpenClaw uses NVIDIA’s hosted OpenAI-compatible endpoint, not a local GPU/model.

So with:

openclaw models set nvidia/nvidia/nemotron-3-super-120b-a12b

your requests go to NVIDIA’s API using NVIDIA_API_KEY. Local would be something like Ollama/llama.cpp/vLLM instead.

reef kindle Apr 27, 2026, 4:23 PM

#

i've got an nvidia moonshotai api key with kimi 2.5. Is it a better model?

velvet spokeBOT Apr 27, 2026, 4:24 PM

#

Yes, it may be better for coding / reasoning / agent-style tasks, but I wouldn’t call it universally “better” without testing your workflow.

If your key is from NVIDIA build.nvidia.com, use the NVIDIA provider model ref:

openclaw models set nvidia/moonshotai/kimi-k2.5

OpenClaw’s NVIDIA docs list Kimi K2.5 as a built-in NVIDIA model, and NVIDIA describes Kimi K2.5 as strong for reasoning, coding, and agentic workflows. (docs.openclaw.ai)

My practical take: try Kimi K2.5 first if you’re doing coding/automation; keep Nemotron 3 Super 120B as the “boring default” fallback if Kimi acts flaky. For live model-quality opinions, #models is the better channel because this changes fast.

OpenClaw

NVIDIA - OpenClaw

reef kindle Apr 27, 2026, 4:25 PM

#

where should I put the api key?

velvet spokeBOT Apr 27, 2026, 4:26 PM

#

Put it in an environment variable named:

NVIDIA_API_KEY

Best persistent place: create/edit:

~/.openclaw/.env

and add:

NVIDIA_API_KEY=nvapi-...

#

Then restart the gateway:

openclaw gateway restart

After that set the model, for Kimi 2.5 via NVIDIA:

openclaw models set nvidia/moonshotai/kimi-k2.5

You can also put it in ~/.openclaw/openclaw.json under env, but .env is cleaner for secrets:

#

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

Don’t paste the key here in Discord 🙂 Docs: https://docs.openclaw.ai/providers/nvidia

reef kindle Apr 27, 2026, 6:54 PM

#

how to do allow this: channels.discord.threadBindings.spawnSubagentSessions=t

velvet spokeBOT Apr 27, 2026, 6:55 PM

#

Use boolean true — not t.

Run:

openclaw config set channels.discord.threadBindings.enabled true --strict-json
openclaw config set channels.discord.threadBindings.spawnSubagentSessions true --strict-json
openclaw gateway restart

If you run from source, prefix with pnpm:

pnpm openclaw config set channels.discord.threadBindings.enabled true --strict-json
pnpm openclaw config set channels.discord.threadBindings.spawnSubagentSessions true --strict-json
pnpm openclaw gateway restart

#

Equivalent in ~/.openclaw/openclaw.json:

{
  channels: {
    discord: {
      threadBindings: {
        enabled: true,
        spawnSubagentSessions: true,
      },
    },
  },
}

That enables Discord thread-bound subagent sessions. For ACP/Codex-style thread spawns, the separate key is spawnAcpSessions: true.

#

Docs: https://docs.openclaw.ai/channels/discord

reef kindle Apr 27, 2026, 7:09 PM

#

this is a subagent config:

❌ Can't post the intro message — bot lacks channel permissions

What's needed:
The Discord bot token needs permissions for channel 1456609488202105005 in your Openclaw server. Likely needs:
Read access to the channel
Send message permissions
Proper server invite with bot scope

Once you fix the permissions, Elena will automatically send her intro and start monitoring for showcase highlights and event requests. Want me to help troubleshoot the Discord bot setup

velvet spokeBOT Apr 27, 2026, 7:10 PM

#

That error is Discord permissions, not really the subagent config.

If this is your own Discord server, fix it by giving the bot access to channel 1456609488202105005:

Discord server → Server Settings → Roles
Select the bot’s role
Enable at least:
- View Channels
- Send Messages
- Read Message History
- Send Messages in Threads if it posts inside threads
- optionally Embed Links / Attach Files
Also check the specific channel/category permissions, because channel overrides can block the bot even if the role is okay.
Restart OpenClaw:

#

openclaw gateway restart

Then test:

openclaw channels status --probe

If this is the public OpenClaw Discord server, don’t try to add/use a personal bot there — use your own private server for that setup.

Also make sure your bot invite included both scopes:

bot

#

applications.commands

Docs: https://docs.openclaw.ai/channels/discord

reef kindle Apr 27, 2026, 9:50 PM

#

openclaw session list
Config warnings:
plugins.allow: plugin not found: session (stale config entry ignored; remove it from plugins config)
error: unknown command 'session'

velvet spokeBOT Apr 27, 2026, 9:50 PM

#

Two separate things there:

The command is plural — there is no openclaw session list.

Use:

openclaw sessions

or:

openclaw sessions --all-agents
openclaw sessions --json

#

This warning:

plugins.allow: plugin not found: session

means your config has a stale plugin allowlist entry named session. session is not a plugin id.

Safest cleanup:

openclaw doctor --fix

Or inspect manually:

#Nvidia API