feat(agents): support per-agent local model lean mode by dutifulbob · Pull Request #84073 · openclaw/openclaw

dutifulbob · 2026-05-19T09:31:42Z

Summary

Add agents.list[].experimental.localModelLean so lean local-model mode can be enabled or disabled per configured agent.
Resolve lean mode from the explicit agent, agent-scoped session key, or configured default agent, then trim browser, cron, and message from normal, bundled, and catalog-compacted tool surfaces.
Update schema/help/labels/docs and add coverage for global inheritance, per-agent opt-out, default-agent resolution, and legacy session keys.

Verification

node scripts/run-vitest.mjs src/agents/local-model-lean.test.ts src/agents/pi-tools-agent-config.test.ts src/agents/pi-tools.model-provider-collision.test.ts src/agents/agent-scope.test.ts src/config/config.schema-regressions.test.ts
pnpm build
node dist/index.js config validate --json | jq '{valid,path,issues:(.issues // [])}'
Local gateway restart plus Discord Looper probe to the LM Studio/Gemma route.

Real behavior proof (required for external PRs)

Behavior or issue addressed: Per-agent local-model lean mode can now be configured for one agent, such as an LM Studio/Gemma test agent, without changing the main agent.
Real environment tested: Local built gateway with Discord lm-studio-test via Looper, routed to the LM Studio Gemma agent. The installed config used agents.list[].experimental.localModelLean: true for the lmstudio agent; this proof did not use agents.list[].tools.deny.
Exact steps or command run after this patch: Built the branch with pnpm build, validated config with node dist/index.js config validate --json, ran node dist/index.js doctor --fix --non-interactive to remove stale external bundled-plugin install records, restarted the user gateway service, and sent a Looper probe to Discord lm-studio-test.
Evidence after fix (screenshot, recording, terminal capture, console output, redacted runtime log, linked artifact, or copied live output): Redacted runtime log excerpt copied after Looper message 1506276275742834851:

{
  "runId": "0f0b16eb-a5d6-4643-a06f-144d56583a78",
  "agentId": "lmstudio",
  "messageProvider": "discord",
  "localModelLean": true,
  "toolCount": 33,
  "clientToolCount": 0,
  "compiledToolCount": 33,
  "compiledToolSampleExcludes": ["browser", "cron", "message"],
  "assistantTexts": ["pong"]
}

Observed result after fix: The LM Studio run recorded localModelLean: true, omitted browser, cron, and message from the compiled tool surface, and completed successfully with pong. Per-agent false is still covered by tests and opts out of inherited global lean mode.
What was not tested: I did not run a full cross-platform or packaged release install. The live proof used the local built gateway with the Discord bundled plugin loaded from this branch.
Before evidence (optional but encouraged): Before running doctor --fix, the local Discord plugin resolved through a stale external bundled-plugin install record, so the live route did not see this branch's per-agent schema.

clawsweeper · 2026-05-19T09:32:55Z

Codex review: passed.

Workflow note: Future ClawSweeper reviews update this same comment in place.

How this review workflow works

ClawSweeper keeps one durable marker-backed review comment per issue or PR.
Re-runs edit this comment so the latest verdict, findings, and automation markers stay together instead of adding duplicate bot comments.
A fresh review can be triggered by eligible @clawsweeper re-review comments, exact-item GitHub events, scheduled/background review runs, or manual workflow dispatch.
PR/issue authors and users with repository write access can comment @clawsweeper re-review or @clawsweeper re-run on an open PR or issue to request a fresh review only.
Maintainers can also comment @clawsweeper review to request a fresh review only.
Fresh-review commands do not start repair, autofix, rebase, CI repair, or automerge.
Maintainer-only repair and merge flows require explicit commands such as @clawsweeper autofix, @clawsweeper automerge, @clawsweeper fix ci, or @clawsweeper address review.
Maintainers can comment @clawsweeper explain to ask for more context, or @clawsweeper stop to stop active automation.

Summary
The branch adds per-agent agents.list[].experimental.localModelLean config and applies lean tool filtering through agent, session, and default-agent resolution.

Reproducibility: not applicable. this is a feature/config PR rather than a current-main bug report. The changed behavior is supported by source review, focused tests in the branch, and the PR body's redacted live runtime log.

PR rating
Overall: 🐚 platinum hermit
Proof: 🐚 platinum hermit
Patch quality: 🐚 platinum hermit
Summary: Good normal PR quality: the patch is scoped, proof is sufficient, and no blocking correctness or security issue was found.

Rank-up moves:

none

What the crustacean ranks mean

🦀 challenger crab: rare, exceptional readiness with strong proof, clean implementation, and convincing validation.
🦞 diamond lobster: very strong readiness with only minor maintainer review expected.
🐚 platinum hermit: good normal PR, likely mergeable with ordinary maintainer review.
🦐 gold shrimp: useful signal, but proof or patch confidence is still limited.
🦪 silver shellfish: thin signal; proof, validation, or implementation needs work.
🧂 unranked krab: not merge-ready because proof is missing/unusable or there are serious correctness or safety concerns.
🌊 off-meta tidepool: rating does not apply to this item.

Shiny media proof means a screenshot, video, or linked artifact directly shows the changed behavior. Runtime, network, CSP, and security claims still need visible diagnostics.

PR egg
✨ Hatched: 🥚 common Gilded Shellbean

       /\  .---.  /\         
      /  \/     \/  \        
     /   ( -   - )   \       
    |       ._.       |      
    |   /|  ===  |\   |      
     \  \|______/|/  /       
      '._  `--'  _.'         
         '-.__.-'            
       _/|_|  |_|\_          
      /__|      |__\         
       .-----------.         
      '-------------'

Rarity: 🥚 common.
Trait: collects tiny proofs.
Share on X: post this hatch
Copy: My PR egg hatched a 🥚 common Gilded Shellbean in ClawSweeper.

What is this egg doing here?

Eggs appear after the PR passes real-behavior proof. It is here for vibes, not verdicts: it does not change labels, ratings, merge decisions, or automation.
The shell reacts to review momentum: open follow-up work warms it up, re-review makes it wobble, and a clean final review lets it hatch.
How to hatch it: reach status: 👀 ready for maintainer look or status: 🚀 automerge armed; that usually means sufficient real-behavior proof, no blocking P0/P1/P2 findings, no security attention needed, and clean correctness.
The hatch is seeded from this repository and PR number, so the same PR keeps the same creature; the reviewed head SHA can only change safe visual details.
Rarity is just collectible sparkle: 🥚 common, 🌱 uncommon, 💎 rare, ✨ glimmer, and 🌈 legendary.

Real behavior proof
Sufficient (logs): The PR body supplies redacted runtime log proof from a local built gateway and Discord Looper route using the new per-agent key, showing the expected lean flag, tool exclusions, and successful response.

Next step before merge
No repair lane is needed; the PR is already on the automerge path and exact-head checks can gate the final merge.

Security
Cleared: The diff changes agent/config/docs/tests only and does not add dependencies, workflows, permissions, downloads, lifecycle scripts, secrets handling, or package execution paths.

Review details

Best possible solution:

Land the opt-in per-agent experimental preset through the exact-head automerge and CI gates while keeping stable manual tool policy on tools.allow and tools.deny.

Do we have a high-confidence way to reproduce the issue?

Not applicable; this is a feature/config PR rather than a current-main bug report. The changed behavior is supported by source review, focused tests in the branch, and the PR body's redacted live runtime log.

Is this the best way to solve the issue?

Yes; the branch extends the existing experimental lean-mode contract through the established agent config resolver and preserves current default behavior for existing users.

Label justifications:

P2: This is a normal focused agent/config feature with opt-in behavior and tests across the affected resolver and tool-policy surfaces.

What I checked:

Current main only has global lean filtering: Current main filters browser, cron, and message only from agents.defaults.experimental.localModelLean; the per-agent runtime schema does not yet expose agents.list[].experimental. (src/agents/pi-tools.ts:241, b7ba7c3f2a1e)
Current main lacks the per-agent schema key: Current main's agent entry schema goes from bootstrap settings to skills without a per-agent experimental object. (src/config/zod-schema.agent-runtime.ts:971, b7ba7c3f2a1e)
PR resolves lean mode from explicit agent, session, or default agent: The new helper normalizes explicit agent ids, parses agent session keys, falls back to the configured default agent, and filters the lean deny set only when the resolved experimental flag is true. (src/agents/local-model-lean.ts:13, 1f9a9554da52)
PR merges per-agent experimental flags with defaults: Resolved agent config now carries experimental and merges omitted per-agent fields over agents.defaults.experimental, which supports per-agent opt-out of inherited lean mode. (src/agents/agent-scope-config.ts:132, 1f9a9554da52)
PR threads agent/session context into tool policy: applyModelProviderToolPolicy now receives agentId and sessionKey and delegates lean filtering to the shared helper before provider-specific filtering. (src/agents/pi-tools.ts:244, 1f9a9554da52)
PR covers bundled and catalog-compacted tool surfaces: The embedded runner filters the combined core and bundled tools before cataloging and filters the catalog result again, so lean mode also applies when tool search or code mode changes the visible surface. (src/agents/pi-embedded-runner/run/attempt.ts:1695, 1f9a9554da52)

Likely related people:

ImLukeF: Commit 4015138 introduced the shipped global lean local-model mode across src/agents/pi-tools.ts, config schema/types, docs, tests, and changelog. (role: original lean-mode feature owner; confidence: high; commits: 4015138df93e, 0abe64a4ff70; files: src/agents/pi-tools.ts, src/config/types.agent-defaults.ts, src/config/zod-schema.agent-defaults.ts)
dutifulbob: Beyond this PR, prior merged main history from Bob touched embedded runner/tool-call behavior and session runtime surfaces adjacent to this change. (role: recent agent-runtime contributor; confidence: medium; commits: 74f2c4a56b4f, 8c7f17b95382, 380de88a61de; files: src/agents/pi-embedded-runner/run/attempt.ts, src/config/zod-schema.agent-runtime.ts, src/agents/tools/session-status.runtime.ts)
osolmaz: Prior related agent-runtime commits by Bob were reviewed by @osolmaz, and @osolmaz explicitly requested ClawSweeper automerge for this PR. (role: reviewer and automerge requester; confidence: medium; commits: 74f2c4a56b4f, 8c7f17b95382, 380de88a61de; files: src/agents/pi-embedded-runner/run/attempt.ts, src/config/zod-schema.agent-runtime.ts)

Codex review notes: model gpt-5.5, reasoning high; reviewed against b7ba7c3f2a1e.

dutifulbob · 2026-05-19T12:47:04Z

Updated the PR body with fresh real-environment proof using the new per-agent key directly.

Key points:

The local installed config used agents.list[].experimental.localModelLean: true for the lmstudio agent.
This proof did not use agents.list[].tools.deny.
Discord was loading a stale external bundled-plugin install record, so I ran node dist/index.js doctor --fix --non-interactive; after that the gateway loaded Discord from this branch's bundled plugin runtime.
Looper message 1506276275742834851 produced LM Studio run 0f0b16eb-a5d6-4643-a06f-144d56583a78.
The run recorded localModelLean: true, toolCount: 33, omitted browser, cron, and message, and completed with pong.
Verification also passed with focused tests, pnpm build, and config validation.

Rationale for keeping this separate from tools.deny: tools.deny is the lower-level manual policy surface, while experimental.localModelLean is a named local-model preset that can stay consistent as the lean surface evolves and can be assigned per agent for LM Studio/Ollama/etc. experiments without copying a deny list into every agent.

@clawsweeper re-review

clawsweeper · 2026-05-19T12:48:49Z

🦞🧹
ClawSweeper re-review requested.

I asked ClawSweeper to review this item again.
Action: item re-review queued (workflow sweep.yml, event repository_dispatch).
Result: the existing ClawSweeper review comment will be edited in place when the review finishes.

Re-review progress:

State: Complete
Detail: The targeted re-review finished, the durable review comment was updated, and the synced verdict was routed.
Run: https://github.com/openclaw/clawsweeper/actions/runs/26098201343
Updated: 2026-05-19T12:58:01.513Z

osolmaz · 2026-05-19T13:27:55Z

@clawsweeper automerge

clawsweeper · 2026-05-19T13:29:49Z

🦞✅
ClawSweeper merged this PR after the passing review.

Source: clawsweeper[bot]
Feedback: structured ClawSweeper verdict: pass (sha=1f9a9554da521f35d6127f729e41401a272b68c0)
Merge status: merged by ClawSweeper automerge
Merged at: 2026-05-19T14:11:39Z
Merge commit: 13c97c5a8d04

What merged:

The branch adds per-agent agents.list[].experimental.localModelLean config and applies lean tool filtering through agent, session, and default-agent resolution.
Reproducibility: not applicable. this is a feature/config PR rather than a current-main bug report. The chan ... or is supported by source review, focused tests in the branch, and the PR body's redacted live runtime log.

Automerge notes:

PR branch already contained follow-up commit before automerge: feat(agents): support per-agent local model lean mode
PR branch already contained follow-up commit before automerge: fix(clawsweeper): address review for automerge-openclaw-openclaw-8407…

The automerge loop is complete.

Automerge progress:

2026-05-19 13:57:31 UTC review queued 345813e2583f (queued)

2026-05-19 13:36:42 UTC review passed 345813e2583f (structured ClawSweeper verdict: pass (sha=345813e2583fa74100970ce8fc8b4dca21733...)

2026-05-19 13:37:03 UTC repair queued 345813e2583f (autonomous) Run: https://github.com/openclaw/clawsweeper/actions/runs/26100834742

2026-05-19 13:43:59 UTC repair started (running) in 1s Run: https://github.com/openclaw/clawsweeper/actions/runs/26100834742 automerge-openclaw-openclaw-84073

2026-05-19 13:44:18 UTC validation plan (passed) in 19s Run: https://github.com/openclaw/clawsweeper/actions/runs/26100834742 pnpm check:changed; pnpm lint; pnpm check:test-types

2026-05-19 13:44:33 UTC Codex write preflight (passed) in 35s Run: https://github.com/openclaw/clawsweeper/actions/runs/26100834742 danger-full-access

2026-05-19 13:51:15 UTC Codex edit 1 bd7dba62a74c (complete) in 7m 16s Run: https://github.com/openclaw/clawsweeper/actions/runs/26100834742 exit 0

2026-05-19 14:02:21 UTC validation and review 1 1f9a9554da52 (complete) in 18m 23s Run: https://github.com/openclaw/clawsweeper/actions/runs/26100834742 already-current

2026-05-19 14:02:29 UTC repair completed 1f9a9554da52 (branch updated) in 18m 31s Run: https://github.com/openclaw/clawsweeper/actions/runs/26100834742 initial automerge rebase is delegated to Codex repair

2026-05-19 14:02:29 UTC review queued 1f9a9554da52 (after repair)

2026-05-19 14:11:34 UTC automerge wait 1f9a9554da52 (ready) in 27m 36s Run: https://github.com/openclaw/clawsweeper/actions/runs/26100834742 checks and exact-head review are ready

2026-05-19 14:11:36 UTC repair finished 1f9a9554da52 (pushed) in 27m 38s Run: https://github.com/openclaw/clawsweeper/actions/runs/26100834742 repair_contributor_branch

2026-05-19 14:11:26 UTC review passed 1f9a9554da52 (structured ClawSweeper verdict: pass (sha=1f9a9554da521f35d6127f729e41401a272b6...)

2026-05-19 14:11:42 UTC merged 1f9a9554da52 (merged by ClawSweeper automerge)

…3 (1)

Summary: - The branch adds per-agent `agents.list[].experimental.localModelLean` config and applies lean tool filtering through agent, session, and default-agent resolution. - Reproducibility: not applicable. this is a feature/config PR rather than a current-main bug report. The chan ... or is supported by source review, focused tests in the branch, and the PR body's redacted live runtime log. Automerge notes: - PR branch already contained follow-up commit before automerge: feat(agents): support per-agent local model lean mode - PR branch already contained follow-up commit before automerge: fix(clawsweeper): address review for automerge-openclaw-openclaw-8407… Validation: - ClawSweeper review passed for head 1f9a9554da521f35d6127f729e41401a272b68c0. - Required merge gates passed before the squash merge. Prepared head SHA: 1f9a9554da521f35d6127f729e41401a272b68c0 Review: openclaw/openclaw#84073 (comment) Co-authored-by: Bob <dutifulbob@gmail.com> Co-authored-by: clawsweeper <274271284+clawsweeper[bot]@users.noreply.github.com> Co-authored-by: clawsweeper[bot] <274271284+clawsweeper[bot]@users.noreply.github.com> Approved-by: osolmaz Co-authored-by: osolmaz <2453968+osolmaz@users.noreply.github.com>

Summary: - The branch adds per-agent `agents.list[].experimental.localModelLean` config and applies lean tool filtering through agent, session, and default-agent resolution. - Reproducibility: not applicable. this is a feature/config PR rather than a current-main bug report. The chan ... or is supported by source review, focused tests in the branch, and the PR body's redacted live runtime log. Automerge notes: - PR branch already contained follow-up commit before automerge: feat(agents): support per-agent local model lean mode - PR branch already contained follow-up commit before automerge: fix(clawsweeper): address review for automerge-openclaw-openclaw-8407… Validation: - ClawSweeper review passed for head 1f9a955. - Required merge gates passed before the squash merge. Prepared head SHA: 1f9a955 Review: openclaw#84073 (comment) Co-authored-by: Bob <dutifulbob@gmail.com> Co-authored-by: clawsweeper <274271284+clawsweeper[bot]@users.noreply.github.com> Co-authored-by: clawsweeper[bot] <274271284+clawsweeper[bot]@users.noreply.github.com> Approved-by: osolmaz Co-authored-by: osolmaz <2453968+osolmaz@users.noreply.github.com>

openclaw-barnacle Bot added docs Improvements or additions to documentation agents Agent runtime and tooling size: M triage: needs-real-behavior-proof Candidate: external PR needs after-fix proof from a real setup. labels May 19, 2026

dutifulbob force-pushed the feat/per-agent-local-model-lean branch from 0f39f13 to 345813e Compare May 19, 2026 10:10

openclaw-barnacle Bot added proof: supplied External PR includes structured after-fix real behavior proof. and removed triage: needs-real-behavior-proof Candidate: external PR needs after-fix proof from a real setup. labels May 19, 2026

clawsweeper Bot added the clawsweeper:automerge Maintainer opted this PR into bounded ClawSweeper-reviewed automerge label May 19, 2026

clawsweeper Bot added status: 🚀 automerge armed This PR is in ClawSweeper's automerge lane. and removed status: 👀 ready for maintainer look ClawSweeper has no concrete contributor-facing blocker left for this PR. labels May 19, 2026

feat(agents): support per-agent local model lean mode

bd7dba6

clawsweeper Bot added 2 commits May 19, 2026 13:51

feat(agents): support per-agent local model lean mode

2ce3828

fix(clawsweeper): address review for automerge-openclaw-openclaw-8407…

1f9a955

…3 (1)

clawsweeper Bot force-pushed the feat/per-agent-local-model-lean branch from 345813e to 1f9a955 Compare May 19, 2026 14:02

clawsweeper Bot merged commit 13c97c5 into openclaw:main May 19, 2026
97 checks passed

osolmaz deleted the feat/per-agent-local-model-lean branch May 19, 2026 14:32

github-actions Bot mentioned this pull request May 19, 2026

📡 Upstream Digest — 2026-05-19 14:40 UTC curtismercier/openclaw-mods#897

Open

This was referenced May 21, 2026

[Feature]: Per-agent cost budget enforcement at the gateway level #42475

Open

fix(update): prune stale local bundled plugin shadows #84863

Merged

clawsweeper Bot mentioned this pull request May 28, 2026

fix(auto-reply): preserve post-compaction failure context in user-facing recovery message (#67750) #87691

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

feat(agents): support per-agent local model lean mode#84073

feat(agents): support per-agent local model lean mode#84073
clawsweeper[bot] merged 3 commits into
openclaw:mainfrom
osolmaz:feat/per-agent-local-model-lean

dutifulbob commented May 19, 2026 •

edited

Loading

Uh oh!

clawsweeper Bot commented May 19, 2026 •

edited

Loading

Uh oh!

dutifulbob commented May 19, 2026

Uh oh!

clawsweeper Bot commented May 19, 2026 •

edited

Loading

Uh oh!

osolmaz commented May 19, 2026

Uh oh!

clawsweeper Bot commented May 19, 2026 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

dutifulbob commented May 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Verification

Real behavior proof (required for external PRs)

Uh oh!

clawsweeper Bot commented May 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

dutifulbob commented May 19, 2026

Uh oh!

clawsweeper Bot commented May 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

osolmaz commented May 19, 2026

Uh oh!

clawsweeper Bot commented May 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

dutifulbob commented May 19, 2026 •

edited

Loading

clawsweeper Bot commented May 19, 2026 •

edited

Loading

clawsweeper Bot commented May 19, 2026 •

edited

Loading

clawsweeper Bot commented May 19, 2026 •

edited

Loading