PR #71486: fix(model): preserve LM Studio "@" quant suffixes in model name resolution

• Repository: openclaw/openclaw
• Author: Bartok9
• State: closed | merged: True
• Link: https://github.com/openclaw/openclaw/pull/71486

Description (problem / solution / changelog)

Summary

Fixes #71474 — LM Studio model names containing @ for quantization variants (e.g. qwen3.6-27b@iq3_xxs) were incorrectly truncated during model resolution.

Problem

stripModelProfileSuffix() in providers.ts naively stripped everything after the first @, discarding quant tags that LM Studio uses to distinguish quantization levels:

lmstudio/qwen3.6-27b@iq3_xxs  →  lmstudio/qwen3.6-27b  (❌ model not found)

This caused:

1. /model lmstudio/qwen3.6-27b@iq3_xxs → model not allowed: lmstudio/qwen3.6-27b
2. API requests sent truncated model name → LM Studio picked a random quant instead of the configured one

Fix

providers.ts: Replace the naive indexOf("@") strip with splitTrailingAuthProfile(), which already has quant-aware logic (preserves @q8_0, @4bit, etc. while still allowing @profile auth suffixes).

model-ref-profile.ts: Extend the quant-suffix regex from q\d+ to i?q\d+ to also match importance-quantization tags (iq3_xxs, iq4_xs) used by llama.cpp / LM Studio.

Changes

Testing

All 15 model-ref-profile tests pass, including 3 new ones:

• @iq3_xxs preserved in model id ✅
• @iq4_xs preserved in model id ✅
• @iq3_xxs@work correctly splits quant from auth profile ✅

Minimal, surgical change — only touches the model name parsing path.

Changed files

• CHANGELOG.md (modified, +3/-0)
• src/agents/model-ref-profile.test.ts (modified, +16/-0)
• src/agents/model-ref-profile.ts (modified, +5/-2)
• src/agents/model-selection.test.ts (modified, +24/-0)
• src/auto-reply/model.test.ts (modified, +14/-0)
• src/plugins/providers.test.ts (modified, +46/-0)
• src/plugins/providers.ts (modified, +2/-3)

PR #71488: fix(agents): preserve LM Studio IQ quant model refs

• Repository: openclaw/openclaw
• Author: chen-zhang-cs-code
• State: closed | merged: False
• Link: https://github.com/openclaw/openclaw/pull/71488

Description (problem / solution / changelog)

Summary

• Problem: LM Studio model ids such as qwen3.6-27b@iq3_xxs were parsed as qwen3.6-27b plus an auth profile suffix.
• Why it matters: /model lmstudio/...@iq* could fail allowlist validation, and configured LM Studio IQ quant variants could lose the exact quant id before inference.
• What changed: Treat @iq* local quant suffixes like the existing @q* / @4bit suffixes and keep them in the model id. Added regression coverage for direct profile splitting and model selection resolution.
• What did NOT change (scope boundary): No LM Studio runtime transport, preload, auth-profile, or catalog discovery behavior changed.

Change Type (select all)

• Bug fix
• Feature
• Refactor required for the fix
• Docs
• Security hardening
• Chore/infra

Scope (select all touched areas)

• Gateway / orchestration
• Skills / tool execution
• Auth / tokens
• Memory / storage
• Integrations
• API / contracts
• UI / DX
• CI/CD / infra

Linked Issue/PR

• Closes #71474
• This PR fixes a bug or regression

Root Cause (if applicable)

• Root cause: splitTrailingAuthProfile already protected common local quant suffixes like @q8_0 and @4bit, but did not include LM Studio IQ quant suffixes such as @iq3_xxs.
• Missing detection / guardrail: Existing tests covered @q* and bit-depth suffixes, but not @iq* model ids from LM Studio.
• Contributing context (if known): LM Studio can expose multiple quants for the same base model by adding IQ suffixes to the model key.

Regression Test Plan (if applicable)

• Coverage level that should have caught this:
  • Unit test
  • Seam / integration test
  • End-to-end test
  • Existing coverage already sufficient
• Target test or file: src/agents/model-ref-profile.test.ts, src/agents/model-selection.test.ts
• Scenario the test should lock in: lmstudio/qwen3.6-27b@iq3_xxs resolves to provider lmstudio and model qwen3.6-27b@iq3_xxs, not a truncated model plus profile suffix.
• Why this is the smallest reliable guardrail: The truncation happens in the shared model-ref/profile parser before LM Studio runtime code sees the selected model.
• Existing test that already covers this (if any): Existing tests covered @q8_0, @4bit, and profile suffixes but not @iq*.
• If no new test is added, why not: N/A

User-visible / Behavior Changes

LM Studio users can select and run IQ quant model ids containing @iq*, for example lmstudio/qwen3.6-27b@iq3_xxs, without the quant suffix being stripped as an auth profile.

Diagram (if applicable)

Before:
/model lmstudio/qwen3.6-27b@iq3_xxs -> lmstudio/qwen3.6-27b + profile iq3_xxs -> model not allowed

After:
/model lmstudio/qwen3.6-27b@iq3_xxs -> lmstudio/qwen3.6-27b@iq3_xxs -> exact configured model

Security Impact (required)

• New permissions/capabilities? (Yes/No) No
• Secrets/tokens handling changed? (Yes/No) No
• New/changed network calls? (Yes/No) No
• Command/tool execution surface changed? (Yes/No) No
• Data access scope changed? (Yes/No) No
• If any Yes, explain risk + mitigation: N/A

Repro + Verification

Environment

• OS: Windows 11 local checkout
• Runtime/container: Node test runner through repo scripts
• Model/provider: LM Studio model refs with @iq* suffixes
• Integration/channel (if any): /model / shared agent model selection
• Relevant config (redacted): N/A

Steps

1. Parse lmstudio/qwen3.6-27b@iq3_xxs through shared model selection.
2. Parse lmstudio/qwen3.6-27b@iq4_xs@work through auth-profile splitting.
3. Verify the first @iq* suffix stays in the model id and a second suffix can still be used as an auth profile.

Expected

• @iq3_xxs and @iq4_xs remain part of the LM Studio model id.
• A second suffix such as @work still resolves as the auth profile override.

Actual

• Covered by the new regression tests.

Evidence

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

pnpm test src/agents/model-ref-profile.test.ts src/agents/model-selection.test.ts
# Test Files  1 passed (1), Tests 13 passed (13)
# Test Files  1 passed (1), Tests 90 passed (90)

pnpm exec oxfmt --check --threads=1 src/agents/model-ref-profile.ts src/agents/model-ref-profile.test.ts src/agents/model-selection.test.ts
# All matched files use the correct format.

pnpm check:changed
# exit 0

Human Verification (required)

• Verified scenarios: Shared parser preserves @iq3_xxs; shared model selection resolves lmstudio/qwen3.6-27b@iq3_xxs to the exact model id; second suffix auth profile still works after @iq4_xs.
• Edge cases checked: Existing @q8_0, @4bit, @8bit, dated suffix, and profile suffix tests still pass.
• What you did not verify: I did not run a live LM Studio server.

Review Conversations

• I replied to or resolved every bot review conversation I addressed in this PR.
• I left unresolved only the conversations that still need reviewer or maintainer judgment.

Compatibility / Migration

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

Risks and Mitigations

• Risk: A user could previously intend @iq3_xxs as an auth profile name.
  • Mitigation: This matches the existing local-quant handling for @q*; users can still specify auth profiles as a second suffix, for example lmstudio/model@iq3_xxs@work.

Changed files

• src/agents/model-ref-profile.test.ts (modified, +7/-0)
• src/agents/model-ref-profile.ts (modified, +4/-3)
• src/agents/model-selection.test.ts (modified, +11/-0)

PR #71490: fix(agents): preserve LM Studio @iq* quant suffixes in model refs

• Repository: openclaw/openclaw
• Author: Sanjays2402
• State: closed | merged: False
• Link: https://github.com/openclaw/openclaw/pull/71490

Description (problem / solution / changelog)

Fixes #71474

Problem

LM Studio names imatrix-quant variants like qwen3.6-27b@iq3_xxs (note: iq prefix, not q). splitTrailingAuthProfile only special-cased @q* / <n>bit suffixes, so:

/model lmstudio/qwen3.6-27b@iq3_xxs

was parsed as model="lmstudio/qwen3.6-27b" + profile="iq3_xxs", producing GatewayRequestError: model not allowed: lmstudio/qwen3.6-27b and (when it didn't fail) sending the truncated id to LM Studio, which then picked an arbitrary quant.

Fix

Extend the protected quant pattern to allow an optional i prefix:

- /^(?:q\d+(?:_[a-z0-9]+)*|\d+bit)(?:@|$)/i
+ /^(?:(?:i?q)\d+(?:_[a-z0-9]+)*|\d+bit)(?:@|$)/i

This keeps @iq3_xxs, @iq4_xs, etc. as part of the model id while still allowing an explicit auth profile via a second @ suffix (e.g. lmstudio/foo@iq3_xxs@work).

Tests

Added regression cases in:

• src/agents/model-ref-profile.test.ts
• src/agents/model-selection.test.ts
• src/auto-reply/model.test.ts

Verified the splitter against representative inputs:

Changed files

• src/agents/model-ref-profile.test.ts (modified, +9/-2)
• src/agents/model-ref-profile.ts (modified, +5/-3)
• src/agents/model-selection.test.ts (modified, +22/-0)
• src/auto-reply/model.test.ts (modified, +14/-0)

PR #71635: fix(agents): handle iq* quant suffixes in splitTrailingAuthProfile

• Repository: openclaw/openclaw
• Author: blaspat
• State: closed | merged: False
• Link: https://github.com/openclaw/openclaw/pull/71635

Description (problem / solution / changelog)

Fixes openclaw/openclaw#71474

Problem

LM Studio uses '@' in model names to denote quant variants (e.g. qwen3.6-27b@iq3_xxs, qwen3.6-27b@iq4_xs). OpenClaw's splitTrailingAuthProfile function incorrectly treats iq* suffixes as auth profile delimiters, truncating the model name at '@'.

Result:

• /model lmstudio/qwen3.6-27b@iq3_xxs → error "model not allowed: lmstudio/qwen3.6-27b"
• LM Studio receives truncated name qwen3.6-27b and picks a random quant

Root cause

The quant-suffix regex only covered q* (e.g. q8_0, q4_k_xl) and *bit patterns, but not i* prefixed quants like iq3_xxs, iq4_xs that LM Studio uses.

Fix

Extend the quant-suffix regex in splitTrailingAuthProfile to also match i-prefixed quant suffixes:

- if (/^(?:q\d+(?:_[a-z0-9]+)*|\d+bit)(?:@|$)/i.test(...))
+ if (/^(?:q\d+(?:_[a-z0-9]+)*|i(?:q\d+(?:_[a-z0-9]+)*)?|\d+bit)(?:@|$)/i.test(...))

The i branch handles both standalone i* and iq* quant formats. Auth profiles can still be appended as a second @ suffix (e.g. ...@iq3_xxs@work).

Test additions

• Covers iq3_xxs and iq4_xs LM Studio quant suffixes
• Covers auth profile after @iq* suffix

Changed files

• src/agents/model-ref-profile.test.ts (modified, +16/-0)
• src/agents/model-ref-profile.ts (modified, +4/-3)