fix(venice): harden discovery limits and tool support by vincentkoc · Pull Request #38306 · openclaw/openclaw

vincentkoc · 2026-03-06T20:05:55Z

Summary

Describe the problem and fix in 2–5 bullets:

Problem: Venice discovery and fallback metadata had drifted from the live API, and OpenClaw still wired tools into Venice models that do not support function calling.
Why it matters: default Venice setups could fail with HTTP 400 errors from oversized max_completion_tokens defaults or tools is not supported requests.
What changed: discovery now applies bounded per-model maxCompletionTokens, the static Venice catalog is synced to current live model metadata, and embedded runs/compaction suppress tools for models with compat.supportsTools === false.
What did NOT change (scope boundary): this PR does not change non-Venice provider behavior or broaden tool availability for other providers.

Change Type (select all)

Scope (select all touched areas)

Linked Issue/PR

User-visible / Behavior Changes

Venice onboarding/default model selection now uses safe per-model completion-token limits instead of a stale shared default.
Venice models that do not support function calling no longer receive tool wiring during embedded runs or compaction.
Offline/degraded Venice fallback uses a catalog synced to the current live Venice model list.

Security Impact (required)

New permissions/capabilities? (No)
Secrets/tokens handling changed? (No)
New/changed network calls? (No)
Command/tool execution surface changed? (Yes)
Data access scope changed? (No)
If any Yes, explain risk + mitigation:
Tool execution is reduced, not expanded: Venice models that advertise no function-calling support now have tools suppressed. Discovery still reads the existing public Venice /models endpoint, but API-provided maxCompletionTokens values are normalized and clamped to safe bounds before use.

Repro + Verification

Environment

OS: macOS
Runtime/container: Node 22 / pnpm
Model/provider: Venice
Integration/channel (if any): n/a
Relevant config (redacted): default Venice discovery + embedded runner paths

Steps

Discover Venice models from the live/static catalog.
Start an embedded run with a Venice model that has a lower completion limit or no function-calling support.
Verify model defaults and tool wiring used by the run.

Expected

Venice models use bounded per-model completion-token limits.
Venice models without function-calling support do not receive tools.
Static fallback catalog contains the current Venice model set/metadata.

Actual

Targeted tests and full build pass with the updated Venice discovery, fallback, and tool-gating logic.

Evidence

Attach at least one:

Failing test/log before + passing after
Trace/log snippets
Screenshot/recording
Perf numbers (if relevant)

Human Verification (required)

What you personally verified (not just CI), and how:

Verified scenarios: catalog-known Venice models honoring bounded API maxCompletionTokens, catalog fallback retaining synced metadata, unknown models getting conservative bounded defaults, and tool suppression for non-function-calling Venice models.
Edge cases checked: missing maxCompletionTokens, oversized maxCompletionTokens, unsupported tools on known and unknown Venice models, post-rebase targeted tests.
What you did not verify: live end-to-end Venice API calls with authenticated completions from a real account.

Compatibility / Migration

Backward compatible? (Yes)
Config/env changes? (No)
Migration needed? (No)
If yes, exact upgrade steps:

Failure Recovery (if this breaks)

How to disable/revert this change quickly: revert this PR or disable Venice model discovery and fall back to explicit configured models.
Files/config to restore: src/agents/venice-models.ts, src/agents/pi-embedded-runner/run/attempt.ts, src/agents/pi-embedded-runner/compact.ts
Known bad symptoms reviewers should watch for: Venice models losing tool access unexpectedly, stale catalog entries reappearing, or Venice requests still failing with max_completion_tokens / unsupported-tools 400s.

Risks and Mitigations

List only real risks for this PR. Add/remove entries as needed. If none, write None.

Risk: the static Venice catalog could drift again as Venice changes its public model list.
- Mitigation: live discovery still overrides known entries within safe bounds, and the fallback catalog is now synced to the current live API state.
Risk: some Venice models may actually support tools despite stale metadata.
- Mitigation: tool suppression is only applied when the catalog or live API explicitly reports no function-calling support.

greptile-apps · 2026-03-06T20:13:40Z

Greptile Summary

This PR hardens Venice provider integration in three complementary areas: (1) per-model completion-token limits are now applied from live discovery data and clamped to safe bounds, (2) the static fallback catalog is re-synced to the current Venice model list (including new models, reclassified privacy levels, corrected IDs like claude-opus-4-5/claude-sonnet-4-5, and a fixed name mismatch for minimax-m21), and (3) a new supportsModelTools gate prevents tool/function-calling requests from being sent to Venice models that advertise no function-calling support, in both the embedded runner and the compaction path.

Key observations:

The token-clamping strategy in resolveApiMaxCompletionTokens is intentionally one-directional: live API values can lower but never raise a catalog model's maxTokens. This prevents oversized values from slipping through but means catalog drift must be fixed manually; a comment documenting this would help future maintainers.
deepseek-v3.2 retains reasoning: true from the old catalog. DeepSeek V3.2 is a dense chat/instruct model (not a reasoning/thinking model like R1 variants), so this flag appears incorrect and could cause the runner to apply reasoning-specific request formatting to it.
Minor: the CHANGELOG attribution truncates @vincentkoc as @vincentko.
All other logic — the supportsTools propagation through buildVeniceModelDefinition, discovery, and both execution paths — is correct and well-tested.

Confidence Score: 4/5

Safe to merge; changes reduce failure surface by suppressing unsupported tool calls and clamping token limits, with no broadening of capabilities.
The core logic is correct and well-tested. The one point deducted is for deepseek-v3.2 carrying reasoning: true — a pre-existing but unfixed misclassification that this catalog-sync PR was well-positioned to correct, and which could cause reasoning-specific request formatting to be applied to a non-reasoning model.
src/agents/venice-models.ts — the deepseek-v3.2 reasoning: true flag and the undocumented one-way tool suppression gate deserve a second look.

Comments Outside Diff (1)

src/agents/venice-models.ts, line 163-172 (link)

deepseek-v3.2 is marked as a reasoning model

deepseek-v3.2 has reasoning: true in the catalog, but DeepSeek V3.2 is a dense chat/instruct model, not a reasoning/thinking model (that role belongs to DeepSeek R1 variants). This was present in the previous catalog and is left unchanged here, but the updated catalog sync pass is a natural opportunity to correct it. Incorrect reasoning flags affect how the runner formats requests (e.g. thinking budgets, effort parameters).

Prompt To Fix With AI

This is a comment left during a code review.
Path: src/agents/venice-models.ts
Line: 163-172

Comment:
**`deepseek-v3.2` is marked as a reasoning model**

`deepseek-v3.2` has `reasoning: true` in the catalog, but DeepSeek V3.2 is a dense chat/instruct model, not a reasoning/thinking model (that role belongs to DeepSeek R1 variants). This was present in the previous catalog and is left unchanged here, but the updated catalog sync pass is a natural opportunity to correct it. Incorrect `reasoning` flags affect how the runner formats requests (e.g. thinking budgets, effort parameters).



How can I resolve this? If you propose a fix, please make it concise.

_{Last reviewed commit: c9d1e63}

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: c9d1e63cee

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

vincentkoc · 2026-03-06T23:19:00Z

Addressed the current review items in follow-up commit 94d7813:

capped unknown-model maxTokens against the same 128000 fallback context window used for degraded Venice metadata, with a regression test for missing availableContextTokens
added an inline comment documenting the intentional one-way supportsTools gate for catalog-known models
rechecked the changelog attribution report; the current branch already has the correct handles, so there was no code change needed there

Re-ran:

pnpm exec vitest run src/agents/venice-models.test.ts src/agents/model-tool-support.test.ts src/config/config-misc.test.ts
pnpm build

Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 6d85562052

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

vincentkoc · 2026-03-06T23:34:12Z

Addressed the partial-metadata review point in follow-up commit c6b1832.

Changes:

made Venice model_spec.availableContextTokens and model_spec.capabilities reads tolerant of partial /models records
guarded tool/reasoning/vision capability reads so one malformed entry no longer aborts discovery for the whole payload
added a regression test covering mixed valid + partial discovery payloads

Re-ran:

pnpm exec vitest run src/agents/venice-models.test.ts src/agents/model-tool-support.test.ts src/config/config-misc.test.ts
pnpm build

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: c6b1832406

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

vincentkoc · 2026-03-07T00:05:53Z

Addressed the missing-model_spec regression in follow-up commit 96ea772.

Changes:

made Venice model_spec optional in discovery parsing
guarded maxCompletionTokens, context-window, name, and tool-capability reads behind optional access
added a regression test proving a malformed known-model row without model_spec no longer aborts discovery for the rest of the payload

Re-ran:

pnpm exec vitest run src/agents/venice-models.test.ts src/agents/model-tool-support.test.ts src/config/config-misc.test.ts
pnpm build

@greptile-apps

* Config: add supportsTools compat flag * Agents: add model tool support helper * Venice: sync discovery and fallback metadata * Agents: skip tools for unsupported models * Changelog: note Venice provider hardening * Update CHANGELOG.md * Venice: cap degraded discovery metadata * Apply suggestion from @greptile-apps[bot] Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com> * Venice: tolerate partial discovery capabilities * Venice: tolerate missing discovery specs --------- Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>

@greptile-apps

* Config: add supportsTools compat flag * Agents: add model tool support helper * Venice: sync discovery and fallback metadata * Agents: skip tools for unsupported models * Changelog: note Venice provider hardening * Update CHANGELOG.md * Venice: cap degraded discovery metadata * Apply suggestion from @greptile-apps[bot] Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com> * Venice: tolerate partial discovery capabilities * Venice: tolerate missing discovery specs --------- Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>

@greptile-apps

* Config: add supportsTools compat flag * Agents: add model tool support helper * Venice: sync discovery and fallback metadata * Agents: skip tools for unsupported models * Changelog: note Venice provider hardening * Update CHANGELOG.md * Venice: cap degraded discovery metadata * Apply suggestion from @greptile-apps[bot] Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com> * Venice: tolerate partial discovery capabilities * Venice: tolerate missing discovery specs --------- Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>

(cherry picked from commit 5320ee7) Partial: only types.models.ts (supportsTools field) — gutted files discarded

@greptile-apps

* Config: add supportsTools compat flag * Agents: add model tool support helper * Venice: sync discovery and fallback metadata * Agents: skip tools for unsupported models * Changelog: note Venice provider hardening * Update CHANGELOG.md * Venice: cap degraded discovery metadata * Apply suggestion from @greptile-apps[bot] Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com> * Venice: tolerate partial discovery capabilities * Venice: tolerate missing discovery specs --------- Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>

@greptile-apps

* Config: add supportsTools compat flag * Agents: add model tool support helper * Venice: sync discovery and fallback metadata * Agents: skip tools for unsupported models * Changelog: note Venice provider hardening * Update CHANGELOG.md * Venice: cap degraded discovery metadata * Apply suggestion from @greptile-apps[bot] Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com> * Venice: tolerate partial discovery capabilities * Venice: tolerate missing discovery specs --------- Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>

vincentkoc added 5 commits March 6, 2026 15:04

Config: add supportsTools compat flag

aecf8dd

Agents: add model tool support helper

ba31fe5

Venice: sync discovery and fallback metadata

327f5e5

Agents: skip tools for unsupported models

9f45f0b

Changelog: note Venice provider hardening

32b8be0

vincentkoc self-assigned this Mar 6, 2026

openclaw-barnacle Bot added the agents Agent runtime and tooling label Mar 6, 2026

vincentkoc mentioned this pull request Mar 6, 2026

fix(venice): sync model catalog with live Venice API #38281

Closed

2 tasks

Update CHANGELOG.md

c9d1e63

openclaw-barnacle Bot added size: M maintainer Maintainer-authored PR labels Mar 6, 2026

vincentkoc marked this pull request as ready for review March 6, 2026 20:08

chatgpt-codex-connector Bot reviewed Mar 6, 2026

View reviewed changes

Comment thread src/agents/venice-models.ts Outdated

greptile-apps Bot reviewed Mar 6, 2026

View reviewed changes

Comment thread CHANGELOG.md Outdated

Comment thread src/agents/venice-models.ts

Venice: cap degraded discovery metadata

94d7813

openclaw-barnacle Bot added size: L and removed size: M labels Mar 6, 2026

Apply suggestion from @greptile-apps[bot]

6d85562

Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>

chatgpt-codex-connector Bot reviewed Mar 6, 2026

View reviewed changes

Comment thread src/agents/venice-models.ts Outdated

Venice: tolerate partial discovery capabilities

c6b1832

chatgpt-codex-connector Bot reviewed Mar 6, 2026

View reviewed changes

Comment thread src/agents/venice-models.ts Outdated

Venice: tolerate missing discovery specs

96ea772

vincentkoc merged commit 5320ee7 into main Mar 7, 2026
27 of 28 checks passed

vincentkoc deleted the vincentkoc-code/venice-provider-cluster-fix branch March 7, 2026 00:07

This was referenced Mar 7, 2026

fix: strip tools for Venice models without function calling support #4837

Closed

[Bug]: Venice provider - update model catalog #20156

Closed

This was referenced Mar 7, 2026

fix(#20156): Venice - add missing models and set a new model default #12964

Closed

feat: change Venice default model to Kimi K2.5 #11276

Closed

This was referenced Mar 7, 2026

🦞 OpenClaw 生态日报 2026-03-07 duanyytop/agents-radar#89

Closed

📡 Upstream Digest — 2026-03-07 01:15 UTC curtismercier/openclaw-mods#196

Open

alexyyyander mentioned this pull request Mar 7, 2026

fix/gateway token mismatch 38617 #38676

Closed

modern-sapien mentioned this pull request Mar 15, 2026

[Bug]: Mistral API returns 422 - openai-completions adapter sends max_completion_tokens which Mistral rejects #47079

Closed

alexey-pelykh mentioned this pull request Mar 20, 2026

Cherry-pick issue #842: compaction-config-venice-kilocode-auth-nodes remoteclaw/remoteclaw#1725

Merged

This was referenced Apr 29, 2026

OpenRouter Perplexity Sonar Deep Research falls back in OpenClaw agent/runtime because the model is treated as tool-capable #64175

Open

[Bug]: openai-completions ignores compat.supportsTools=false and still sends tools #74664

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

fix(venice): harden discovery limits and tool support#38306

fix(venice): harden discovery limits and tool support#38306
vincentkoc merged 10 commits intomainfrom
vincentkoc-code/venice-provider-cluster-fix

vincentkoc commented Mar 6, 2026

Uh oh!

greptile-apps Bot commented Mar 6, 2026 •

edited

Loading

Comments Outside Diff (1)

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

vincentkoc commented Mar 6, 2026

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

Uh oh!

vincentkoc commented Mar 6, 2026

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

Uh oh!

vincentkoc commented Mar 7, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Uh oh!

Conversation

vincentkoc commented Mar 6, 2026

Summary

Change Type (select all)

Scope (select all touched areas)

Linked Issue/PR

User-visible / Behavior Changes

Security Impact (required)

Repro + Verification

Environment

Steps

Expected

Actual

Evidence

Human Verification (required)

Compatibility / Migration

Failure Recovery (if this breaks)

Risks and Mitigations

Uh oh!

greptile-apps Bot commented Mar 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Greptile Summary

Confidence Score: 4/5

Comments Outside Diff (1)

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

Uh oh!

Uh oh!

vincentkoc commented Mar 6, 2026

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

vincentkoc commented Mar 6, 2026

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

vincentkoc commented Mar 7, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

greptile-apps Bot commented Mar 6, 2026 •

edited

Loading