#model doesnt recognize tools

1 messages · Page 1 of 1 (latest)

fiery matrix Mar 5, 2026, 8:02 PM

after changing the model, no matter how much i try it says that it doesnt have access to any tools, i checked the config 1000 times, the profile is full. i restarted gateway too
gemini 3 flash

near streamBOT Mar 5, 2026, 8:02 PM

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

hollow forumBOT Mar 5, 2026, 8:05 PM

This usually means the Gateway is no longer offering any tools to the model (so the model correctly says “I don’t have tool access”), and it often happens right after a model/provider switch because of tool policy (allow/deny/profile/byProvider) or sandbox tool policy.

Can you paste these 3 things?

Version (first!)

openclaw --version

Status dumps (safe to share; redacts secrets)

openclaw status --all
openclaw models status

The tool-policy part of your config (only this section is enough):

tools.profile

tools.allow / tools.deny
tools.byProvider (if present)
and if you use sandboxing: tools.sandbox.tools.allow / tools.sandbox.tools.deny
plus any per-agent overrides: agents.list[].tools...

Quick self-checks while you grab that

If you have any allowlist at all, remember it’s restrictive. For example:
- tools.profile: "minimal" ⇒ basically no tools.
- tools.byProvider.google (or google/gemini-…) set to minimal/empty ⇒ tools disappear only on Gemini.
- tools.sandbox.tools.deny: ["*"] (or an empty allowlist) ⇒ tools disappear only in sandboxed chats (common if your Discord threads are “non-main” sessions).
Also check logs for tool-policy warnings:

openclaw logs --follow

Look for lines like “unknown tools in allowlist” / “tool policy” / “allowlist ignored” around the time you send a message.

Once you paste the outputs/config snippet, I can tell you exactly which setting is stripping the tools for Gemini 3 Flash.

fiery matrix Mar 5, 2026, 8:09 PM

version 2026.3.2
Config : ~/.openclaw/openclaw.json
Agent dir : ~/.openclaw/agents/main/agent
Default : google-gemini-cli/gemini-3-pro-preview
Fallbacks (8) : opencode/claude-opus-4-6, opencode/minimax-m2.5-free, qwen-portal/coder-model, qwen-portal/vision-model, openrouter/auto, openrouter/openrouter/auto, google-gemini-cli/gemini-3.1-pro-preview, openrouter/arcee-ai/trinity-large-preview:free
Image model : -
Image fallbacks (0): -
Aliases (3) : Opus -> opencode/claude-opus-4-6, qwen -> qwen-portal/coder-model, OpenRouter -> openrouter/auto
Configured models (9): opencode/claude-opus-4-6, opencode/minimax-m2.5-free, qwen-portal/coder-model, qwen-portal/vision-model, google-gemini-cli/gemini-3-pro-preview, openrouter/auto, openrouter/openrouter/auto, google-gemini-cli/gemini-3.1-pro-preview, openrouter/arcee-ai/trinity-large-preview:free

Auth overview
Auth store : ~/.openclaw/agents/main/agent/auth-profiles.json
Shell env : off
Providers w/ OAuth/tokens (2): google-gemini-cli (1), qwen-portal (1)

google-gemini-cli effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=1 (oauth=1, token=0, api_key=0) | google-gemini-cli:redacted@gmail.com=OAuth (redacted@gmail.com)
opencode effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=1 (oauth=0, token=0, api_key=1) | opencode:default=sk-leBBu...piAhjB33
openrouter effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=1 (oauth=0, token=0, api_key=1) | openrouter:default=sk-or-v1...2887df1b
qwen-portal effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=1 (oauth=1, token=0, api_key=0) | qwen-portal:default=OAuth | models.json=qw...th | source=models.json: ~/.openclaw/agents/main/agent/models.json

OAuth/token status

google-gemini-cli usage: Pro 100% left · Flash 100% left
- google-gemini-cli:redacted@gmail.com (redacted@gmail.com) ok expires in 53m
qwen-portal

OpenClaw status --all

Overview
┌─────────────────┬───────────────────────────────────────────────────────────┐
│ Item │ Value │
├─────────────────┼───────────────────────────────────────────────────────────┤
│ Version │ 2026.3.2 │
│ OS │ linux 6.18.13-arch1-1 (x64) │
│ Node │ 25.7.0 │
│ Config │ ~/.openclaw/openclaw.json │
│ Dashboard │ http://192.168.31.176:18789/ │
│ Tailscale │ off │
│ Channel │ stable (default) │
│ Update │ pnpm · npm latest 2026.3.2 │
│ Gateway │ local · ws://127.0.0.1:18789 (local loopback) · │
│ │ reachable 113ms · auth password │
│ Security │ Run: openclaw security audit --deep │
│ Gateway self │ user-crestlineich8m (192.168.31.176) app 2026.3.2 linux │
│ │ 6.18.13-arch1-1 │
│ Gateway service │ systemd installed · enabled · running (pid 19262) │
│ Node service │ systemd not installed │
│ Agents │ 1 total · 0 bootstrapping · 1 active · 8 sessions │
└─────────────────┴───────────────────────────────────────────────────────────┘

Diagnosis (read-only)

Gateway connection details:
Gateway target: ws://127.0.0.1:18789
Source: local loopback
Config: /home/user/.openclaw/openclaw.json
Bind: lan

✓ Config: /home/user/.openclaw/openclaw.json
✓ Restart sentinel: none
✓ Port 18789
! Tailscale: off · unknown
error: Error: spawn tailscale ENOENT
✓ Skills: 10 eligible · 0 missing · /home/user/.openclaw/workspace
✓ Channel issues (none)

Pasteable debug report. Auth tokens redacted.

hollow forumBOT Mar 5, 2026, 8:11 PM

Your models status output is the key clue: you’re not on the normal Gemini API provider (google/...) — you’re on Gemini CLI:

Default: google-gemini-cli/gemini-3-pro-preview

That’s a different integration path (“shell out to a local gemini binary”) and it’s known to behave differently re: streaming/tool support and schema quirks. (Docs call this out explicitly.)
Source: https://docs.openclaw.ai/help/testing (see the “Gemini API vs Gemini CLI” note)

1) Fastest fix (if you want reliable tool-calling)

Use the Gemini API key provider instead:

Provider: google
Model: google/gemini-3-flash-preview (or pro)

Setup path from docs:

Add GEMINI_API_KEY=... to ~/.openclaw/.env (or run onboarding with the Gemini API key auth choice), then restart the gateway and switch the model.
Source: https://docs.openclaw.ai/concepts/model-providers (Google Gemini section)

2) If you want to debug what’s happening on `google-gemini-cli/...`

Run this and paste the output:

openclaw status --all

And also do:

openclaw logs --follow

Then send a message that would normally use a tool. In the logs, look for:

google tool schema snapshot (it prints toolCount)

If toolCount: 0 → your tool policy is stripping everything (even if you think “profile is full”). Then paste just your tools: { ... } block from ~/.openclaw/openclaw.json.

If toolCount > 0 but the model still claims “no tools” → it’s almost certainly the Gemini CLI path ignoring/struggling with tool calling, and switching to google/... (API) is the pragmatic fix.

If you paste openclaw status --all + a ~20 line log snippet around google tool schema snapshot, I can tell which branch you’re in immediately.

That’s only the overview slice of status --all — the bit we need is further down.

Can you scroll a bit more and paste just these sections from openclaw status --all (they’re safe; secrets are redacted):

Models (shows the effective provider/model + any provider quirks)
Tools / Tool policy (this is the smoking gun: it’ll show whether your allow/deny/profile/byProvider collapsed to “no tools”)
Any Warnings block (often includes “unknown tool names” / “tool allowlist ignored” / etc.)

Also: one log line will answer this immediately

Run:

openclaw logs --follow

Then send a message to your bot that would normally require a tool (anything like “fetch https://example.com” / “read file …” / “list sessions”).

In the logs, copy/paste the line that contains:

google tool schema snapshot

It prints toolCount. If that says toolCount: 0, it’s 100% tool-policy/config. If it’s >0, then the Gemini CLI path is receiving tools but not using them (schema/tool-calling mismatch).

Paste those bits and I’ll tell you the exact fix (usually either a tools.allow/deny issue, or you need to switch from google-gemini-cli/... to the normal google/... Gemini API provider for tool calling).

Got it — that status --all snippet still doesn’t include the tools policy, so we’re missing the one thing that decides whether the model even gets tools.

1) Please paste your `tools` config block

From ~/.openclaw/openclaw.json, copy/paste only the parts that match:

top-level tools: { ... }
and (if present) agents: { defaults: { tools: ... } } or any agents.list[].tools

I’m specifically looking for anything like:

tools.allow / tools.deny
tools.profile
tools.byProvider (this often bites people right after switching to a Google/Gemini provider)

Docs reference (so we’re talking about the same keys): https://docs.openclaw.ai/tools

2) Get one log line: `google tool schema snapshot` (it shows toolCount)

Right now we’re guessing whether the Gateway is sending 0 tools vs “tools are sent but Gemini ignores them”.

Make sure your file logs include info (otherwise this line won’t show up):

// in openclaw.json
{
  logging: { level: "info" },
}

Restart gateway, then:

openclaw logs --follow --json

Send one message that would use a tool, and paste the single log event containing:

"google tool schema snapshot"

It’ll include toolCount.

If toolCount: 0 → config/tool policy issue (almost always tools.byProvider or an allowlist)
If toolCount > 0 → provider/tool-calling mismatch (then the practical fix is usually switching from google-gemini-cli/... to the normal google/... Gemini API provider)

Post the tools block + that one log event and we can pin this down in one round trip.

fiery matrix Mar 5, 2026, 8:17 PM

"tools": {
"profile": "full",
"allow": [
"group:openclaw",
"group:runtime",
"group:fs",
"exec",
"bash",
"process"
],
"loopDetection": {
"enabled": true,
"warningThreshold": 10
},
"message": {
"allowCrossContextSend": true,
"crossContext": {
"allowWithinProvider": true,
"allowAcrossProviders": true
},
"broadcast": {
"enabled": true
}
},
"exec": {
"host": "node",
"security": "full",
"ask": "off"
}
},

hollow forumBOT Mar 5, 2026, 8:19 PM

That tools block looks basically “wide open” — so if you’re still getting “I don’t have access to any tools”, it’s almost certainly not because tools.allow is empty.

Two important gotchas I do see:

A) Your `exec` is configured to run on a node, but you have no node service

You have:

tools: {
  exec: { host: "node", security: "full", ask: "off" }
}

…and your status --all earlier said “Node service: systemd not installed”.

So if what you’re testing is “run a command”, the bot may act like it has no tools because the only tool you care about (exec) can’t run anywhere.

Quick fix (test): change that to run on the gateway host:

tools: { exec: { host: "gateway" } }

(Also: security:"full" + ask:"off" is pretty dangerous with tool-capable models. At least for testing, I’d recommend ask:"on-miss".)

B) You’re using `google-gemini-cli/...` (Gemini CLI), which can be flaky with tool calling

This provider path is explicitly called out in the docs as having “tool support/version skew” quirks vs the normal google/... Gemini API.
Source: https://docs.openclaw.ai/help/testing

The one log line that will tell us which it is

Please run:

Set file logs to info (so the line actually appears):

#model doesnt recognize tools

Quick self-checks while you grab that

1) Fastest fix (if you want reliable tool-calling)

2) If you want to debug what’s happening on google-gemini-cli/...

Also: one log line will answer this immediately

1) Please paste your tools config block

2) Get one log line: google tool schema snapshot (it shows toolCount)

A) Your exec is configured to run on a node, but you have no node service

B) You’re using google-gemini-cli/... (Gemini CLI), which can be flaky with tool calling