You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
- **Lean local model shell access:** keep `exec` directly visible beside the default structured Tool Search controls so coding-tuned local models can use their shell fallback instead of searching for missing domain tools. (#87587) Thanks @vincentkoc.
26
27
- **OAuth refresh contention diagnostics:** keep local lock paths out of user-facing refresh failures and avoid duplicate failure prefixes while preserving structured provider and profile classification. (#83383) Thanks @vincentkoc.
27
28
- **Exec approval prompts:** keep background-disabled fallback warnings out of pending gateway/node approvals and show them only after a command actually runs in the foreground. (#78184) Thanks @vincentkoc.
28
29
- **Direct poll delivery:** route direct and hybrid channel polls through the owning outbound adapter while preserving gateway-mode routing and channel option checks. (#99950) Thanks @NianJiuZst.
Copy file name to clipboardExpand all lines: docs/concepts/experimental-features.md
+2Lines changed: 2 additions & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -29,6 +29,8 @@ Experimental features are opt-in preview surfaces behind explicit flags. They ne
29
29
30
30
If you already tune Tool Search globally, OpenClaw leaves that config alone. Set `tools.toolSearch: false` to opt out of the lean-mode Tool Search default.
31
31
32
+
In structured `tools` mode, lean runs keep `exec` directly visible beside the Tool Search controls so coding-tuned local models can still choose their familiar shell path. This changes schema visibility only: normal tool policy, sandboxing, and exec approvals still apply. Explicit `code` and `directory` modes keep their normal compaction behavior.
33
+
32
34
### Why these tools
33
35
34
36
These tools have the largest descriptions, broadest parameter shapes, or highest chance of distracting a small model from the normal coding and conversation path. On a small-context or stricter OpenAI-compatible backend that is the difference between:
Copy file name to clipboardExpand all lines: docs/gateway/local-models.md
+1-1Lines changed: 1 addition & 1 deletion
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -263,7 +263,7 @@ If the model loads cleanly but full agent turns misbehave, work top-down: confir
263
263
openclaw infer model run --gateway --model <provider/model> --prompt "Reply with exactly: pong" --json
264
264
```
265
265
266
-
3.**Try lean mode** if both probes pass but real agent turns fail with malformed tool calls or oversized prompts: set `agents.defaults.experimental.localModelLean: true`. It drops heavyweight browser, cron, message, media-generation, voice, and PDF tools unless explicitly required, and defaults larger tool catalogs behind structured Tool Search controls. See [Experimental Features -> Local model lean mode](/concepts/experimental-features#local-model-lean-mode) for details and how to confirm it's on.
266
+
3.**Try lean mode** if both probes pass but real agent turns fail with malformed tool calls or oversized prompts: set `agents.defaults.experimental.localModelLean: true`. It drops heavyweight browser, cron, message, media-generation, voice, and PDF tools unless explicitly required, and defaults larger tool catalogs behind structured Tool Search controls while keeping `exec` directly visible. See [Experimental Features -> Local model lean mode](/concepts/experimental-features#local-model-lean-mode) for details and how to confirm it's on.
267
267
268
268
4.**Disable tools entirely as a last resort** by setting `models.providers.<provider>.models[].compat.supportsTools: false` for that model - the agent then runs without tool calls.
0 commit comments