fix: expose ollama thinking profile before activation by yfge · Pull Request #77617 · openclaw/openclaw

yfge · 2026-05-05T01:15:57Z

Summary

expose Ollama's reasoning-capable thinking profile from the lightweight provider-policy artifact
keep /think max available for reasoning-capable Ollama models before the full Ollama plugin runtime activates
add regression coverage for the public artifact path

Changes

add resolveThinkingProfile() to extensions/ollama/provider-policy-api.ts
reuse that resolver from the full Ollama runtime registration
return off/low/medium/high/max for reasoning-capable Ollama models and off otherwise
document the fix in CHANGELOG.md

Real behavior proof

Behavior or issue addressed: ollama/deepseek-v4-pro:cloud rejected /think max before the Ollama plugin runtime activated.
Real environment tested: local OpenClaw build from this PR worktree (openclaw 2026.5.4, head ecde0494faae975180bcda243f80472b810a0f18) with built runtime artifacts generated by pnpm openclaw models --help.

Exact steps or command run after this patch:

node --input-type=module <<'EOF_RUN'
const thinking = await import('./dist/thinking-DtIsL--U.js');
const catalog = [{ provider: 'ollama', id: 'deepseek-v4-pro:cloud', reasoning: true }];
console.log(`levels=${thinking.t('ollama', 'deepseek-v4-pro:cloud', '|', catalog)}`);
console.log(`maxSupported=${thinking.n({ provider: 'ollama', model: 'deepseek-v4-pro:cloud', level: 'max', catalog })}`);
EOF_RUN

Evidence after fix: Terminal capture from the built OpenClaw runtime:
```
levels=off|low|medium|high|max
maxSupported=true
```
Observed result after fix: the startup-safe provider-policy path now advertises max for reasoning-capable Ollama Cloud models before full plugin activation, so the /think max validation path no longer falls back to the base off|minimal|low|medium|high profile.
What was not tested: live Fedora/Ollama Cloud chat round trip.

Testing

pnpm exec oxfmt --check --threads=1 CHANGELOG.md extensions/ollama/index.ts extensions/ollama/provider-policy-api.ts extensions/ollama/provider-policy-api.test.ts
pnpm test extensions/ollama/provider-policy-api.test.ts extensions/ollama/index.test.ts
Testbox pnpm check:changed

Fixes #77612

clawsweeper · 2026-05-05T01:19:09Z

Codex review: needs maintainer review before merge.

Summary
The PR adds an Ollama resolveThinkingProfile export to the provider-policy artifact, regression coverage, and a changelog entry for #77612.

Reproducibility: yes. by source inspection. Current main's fallback path can ask the provider-policy artifact before runtime activation, but Ollama's artifact currently lacks the profile while the runtime hook exposes max for reasoning models.

Next step before merge
No ClawSweeper repair job is needed because this open PR already contains the narrow code, test, and changelog change; the remaining action is normal review and CI.

Security
Cleared: The diff only adds typed provider-policy data, a unit test, and changelog text; it does not touch dependency, workflow, install, secret, or code-execution surfaces.

Review details

Best possible solution:

Land this PR after normal CI or changed-gate validation, keeping the lightweight policy artifact aligned with the full Ollama runtime hook.

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

Yes, by source inspection. Current main's fallback path can ask the provider-policy artifact before runtime activation, but Ollama's artifact currently lacks the profile while the runtime hook exposes max for reasoning models.

Is this the best way to solve the issue?

Yes. Adding the Ollama-owned static profile to extensions/ollama/provider-policy-api.ts is the narrowest maintainable fix and does not require a new core seam or config option.

What I checked:

Current main gap: Current main's Ollama provider-policy artifact exports normalizeConfig only, so the lightweight artifact has no resolveThinkingProfile before runtime activation. (extensions/ollama/provider-policy-api.ts:1, 58c4f9e190bc)
Fallback contract: Thinking profile resolution checks the active provider first, then falls back to resolveBundledProviderPolicySurface(...).resolveThinkingProfile, making the public artifact the relevant pre-activation path. (src/plugins/provider-thinking.ts:76, 58c4f9e190bc)
Runtime Ollama behavior: The full Ollama runtime already exposes off, low, medium, high, and max for reasoning-capable models, and only off otherwise. (extensions/ollama/index.ts:252, 58c4f9e190bc)
Documented behavior: The thinking docs state that Ollama thinking-capable models expose /think low|medium|high|max, with max mapped to native high effort. Public docs: docs/tools/thinking.md. (docs/tools/thinking.md:31, 58c4f9e190bc)
PR implementation: The PR adds a provider-policy resolveThinkingProfile that returns the same reasoning and non-reasoning level sets as the runtime Ollama provider. (extensions/ollama/provider-policy-api.ts:53, b255a8d25158)
PR coverage and changelog: The PR adds a focused provider-policy artifact regression test and a user-facing changelog fix entry for [Bug]: ollama/deepseek-v4-pro:cloud does not allow "max" in v2026.5.3-1 #77612. (extensions/ollama/provider-policy-api.test.ts:61, b255a8d25158)

Likely related people:

vincentkoc: Current blame for the Ollama provider-policy artifact, provider-thinking fallback, and runtime Ollama thinking profile points to Vincent Koc's provider fast-path work; earlier Ollama commits also cover native thinking fallback behavior. (role: recent maintainer and Ollama/provider-policy owner; confidence: high; commits: d253392ea2a3, d6d04f361e1c; files: extensions/ollama/provider-policy-api.ts, src/plugins/provider-thinking.ts, extensions/ollama/index.ts)
Bruce MacDonald: Recent Ollama thinking work changed how OpenClaw sends native think options when thinking is off, adjacent to the same provider-owned thinking behavior. (role: adjacent Ollama thinking maintainer; confidence: medium; commits: 773c57b41882; files: extensions/ollama/src/stream.ts, src/agents/pi-embedded-runner/extra-params.ts)

Remaining risk / open question:

I did not independently run the PR branch tests in this read-only review; merge should still be gated by normal CI or changed-gate proof.

Codex review notes: model gpt-5.5, reasoning high; reviewed against 58c4f9e190bc.

yfge · 2026-05-05T02:50:03Z

Rebased this branch onto current origin/main and resolved the CHANGELOG.md conflict. The code diff remains limited to the Ollama provider-policy artifact/test plus changelog.

Validation after rebase:

pnpm test extensions/ollama/provider-policy-api.test.ts
pnpm exec oxfmt --check --threads=1 CHANGELOG.md extensions/ollama/provider-policy-api.ts extensions/ollama/provider-policy-api.test.ts
pnpm tsgo:extensions && pnpm tsgo:test:extensions

Note: the first typecheck attempt exposed missing local deps (web-tree-sitter); I ran pnpm install and then the extension prod/test typechecks passed.

Fixes openclaw#77612

@yfge

… (thanks @yfge)

obviyus

Verified the Ollama startup-lazy thinking regression: the lightweight Ollama provider-policy artifact now exposes the reasoning-capable profile, so ollama/deepseek-v4-pro:cloud advertises max before the full plugin runtime activates.

Maintainer follow-up rebased onto current main, reused the lightweight resolver from the full Ollama runtime registration, moved the changelog entry to the active Fixes block, and refreshed the PR body's real-behavior proof.

Validation: pnpm test extensions/ollama/provider-policy-api.test.ts extensions/ollama/index.test.ts, targeted oxfmt, Testbox pnpm check:changed, and GitHub Real behavior proof passed. No open review threads remained.

obviyus · 2026-05-05T07:14:21Z

Landed on main.

Landed source commit: ecde049
Merge commit: 5e9258d

Thanks @yfge.

@yfge

…ainer-hardening * origin/main: (843 commits) docs(changelog): relocate openclaw#77046 and openclaw#77280 entries from 2026.5.3 to Unreleased (openclaw#77728) docs: reorder unreleased changelog fix: expose ollama thinking profile before activation (openclaw#77617) (thanks @yfge) fix: expose ollama thinking profile before activation test(gateway): preserve dispatch timers in waiter test(gateway): keep startup context timer live docs: document cache-friendly activity helper ci: install ffmpeg for Mantis media previews fix: avoid impossible device token rotation advice (openclaw#77688) (thanks @Conan-Scott) docs(changelog): note doctor device pairing advice fix fix(doctor): avoid impossible device token rotation advice ci: use Crabbox media previews for Mantis docs: filter maintainer-owned triage noise test: cover GitHub activity helper fix(session-file-repair): drop null-role message entries instead of preserving them (openclaw#77288) fix: prune orphan session artifacts perf: reduce GitHub activity cache misses fix: cache session list model resolution (openclaw#77650) (thanks @ragesaq) ci: embed Mantis desktop previews fix(replay-history): drop trailing stream-error placeholder before provider send (openclaw#77287) ... # Conflicts: # CHANGELOG.md

@yfge

… (thanks @yfge)

openclaw-barnacle Bot added the size: XS label May 5, 2026

This comment was marked as low quality.

Sign in to view

yfge force-pushed the fix/issue-77612 branch from da9e98a to b255a8d Compare May 5, 2026 02:49

obviyus force-pushed the fix/issue-77612 branch from b255a8d to 2e787e6 Compare May 5, 2026 07:00

openclaw-barnacle Bot added the triage: needs-real-behavior-proof Candidate: external PR needs after-fix proof from a real setup. label May 5, 2026

obviyus self-assigned this May 5, 2026

yfge and others added 2 commits May 5, 2026 12:33

fix: expose ollama thinking profile before activation

d9b84c8

Fixes openclaw#77612

fix: expose ollama thinking profile before activation (openclaw#77617)…

ecde049

… (thanks @yfge)

obviyus force-pushed the fix/issue-77612 branch from 2e787e6 to ecde049 Compare May 5, 2026 07:07

obviyus removed the triage: needs-real-behavior-proof Candidate: external PR needs after-fix proof from a real setup. label May 5, 2026

openclaw-barnacle Bot added triage: needs-real-behavior-proof Candidate: external PR needs after-fix proof from a real setup. and removed triage: needs-real-behavior-proof Candidate: external PR needs after-fix proof from a real setup. labels May 5, 2026

obviyus approved these changes May 5, 2026

View reviewed changes

obviyus merged commit 5e9258d into openclaw:main May 5, 2026
98 of 107 checks passed

obviyus mentioned this pull request May 5, 2026

fix(ollama): expose resolveThinkingProfile in provider-policy surface #77652

Closed

github-actions Bot mentioned this pull request May 5, 2026

📡 Upstream Digest — 2026-05-05 09:55 UTC curtismercier/openclaw-mods#773

Open

github-actions Bot pushed a commit to Desicool/openclaw that referenced this pull request May 9, 2026

fix: expose ollama thinking profile before activation (openclaw#77617)…

0b5bdf8

… (thanks @yfge)

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

fix: expose ollama thinking profile before activation#77617

fix: expose ollama thinking profile before activation#77617
obviyus merged 2 commits intoopenclaw:mainfrom
yfge:fix/issue-77612

yfge commented May 5, 2026 •

edited by obviyus

Loading

Uh oh!

clawsweeper Bot commented May 5, 2026 •

edited

Loading

Uh oh!

This comment was marked as low quality.

yfge commented May 5, 2026

Uh oh!

obviyus left a comment

Uh oh!

Uh oh!

obviyus commented May 5, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Conversation

yfge commented May 5, 2026 • edited by obviyus Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Changes

Real behavior proof

Testing

Uh oh!

clawsweeper Bot commented May 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

This comment was marked as low quality.

yfge commented May 5, 2026

Uh oh!

obviyus left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

obviyus commented May 5, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

yfge commented May 5, 2026 •

edited by obviyus

Loading

clawsweeper Bot commented May 5, 2026 •

edited

Loading