fix: Ollama API compatibility — allow bare model names and smart openai/ prefix handling on wire by Copilot · Pull Request #1 · GeorgijeSav/claw-code

Copilot · 2026-05-26T17:44:55Z

@copilot please update the pr description according to the relation ship and the issue which you will find here ultraworkers#3123

Here is the updated PR description reflecting the actual behavior and linking to the upstream issue:

Fixes ultraworkers/claw-code#3123 (related: #3051, #3044)

Ollama/local provider model names like qwen2.5-coder:7b fail in two ways:

Rejected by validate_model_syntax which requires provider/model format.
When prefixed as openai/qwen2.5-coder:7b to satisfy validation, the full string is sent on the wire causing a 404 (model 'openai/qwen2.5-coder:7b' not found).

Changes

validate_model_syntax — allows bare model names (no /) when OPENAI_BASE_URL points to a loopback host (localhost, 127.0.0.1, ::1). Non-local gateways (e.g. OpenRouter) still require the standard provider/model slug so that errors aren't silently masked.
wire_model_for_base_url — base-URL-aware openai/ prefix handling:
- Default OpenAI endpoint (https://api.openai.com/v1): openai/ is stripped (it is only a routing prefix).
- Known-local endpoints (localhost, 127.0.0.1, ::1 — e.g. Ollama, LM Studio, vLLM): openai/ is stripped because local servers use bare model names.
- Custom non-local gateways (OpenRouter, other proxies): the full slug is preserved so the gateway receives the model ID it expects (e.g. openai/gpt-4.1-mini).
Tests — renamed the existing integration test to reflect its actual assertion; added a new test for non-routing slash model IDs being preserved; added unit tests for wire_model_for_base_url and validate_model_syntax local-bypass behavior.

# Before (both fail with Ollama at OPENAI_BASE_URL=http://localhost:11434/v1)
claw --model "qwen2.5-coder:7b" prompt "hi"         # → invalid model syntax
claw --model "openai/qwen2.5-coder:7b" prompt "hi"  # → 404: model 'openai/qwen2.5-coder:7b' not found

# After (both work when OPENAI_BASE_URL=http://localhost:11434/v1)
claw --model "qwen2.5-coder:7b" prompt "hi"         # → sends "qwen2.5-coder:7b" to Ollama
claw --model "openai/qwen2.5-coder:7b" prompt "hi"  # → sends "qwen2.5-coder:7b" to Ollama

# OpenRouter / non-local gateways — behavior unchanged
claw --model "openai/gpt-4.1-mini" prompt "hi"      # → sends "openai/gpt-4.1-mini" to gateway

Fixes two Ollama API compatibility issues (ultraworkers#3123): 1. validate_model_syntax now allows bare model names (e.g. "qwen2.5-coder:7b") when OPENAI_BASE_URL is set, since local providers like Ollama don't use the provider/model naming convention. 2. wire_model_for_base_url always strips the "openai/" routing prefix before sending the model name to the API. This prefix is a claw-code internal routing hint and should never be sent on the wire to any endpoint.

Copilot

Pull request overview

This PR aims to improve compatibility with OpenAI-compatible local endpoints (e.g., Ollama) by allowing bare model names at the CLI layer and by adjusting how model identifiers are sent over the wire for OpenAI-compat requests.

Changes:

Relax validate_model_syntax to accept bare model names when OPENAI_BASE_URL is set.
Change OpenAI-compat request building to strip the openai/ routing prefix before sending the model field.
Update an OpenAI-compat integration test expectation to match the new wire-format behavior.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 4 comments.

File	Description
`rust/crates/rusty-claude-cli/src/main.rs`	Allows bare model names when `OPENAI_BASE_URL` is set; adds a unit test.
`rust/crates/api/src/providers/openai_compat.rs`	Changes `wire_model_for_base_url` to always strip `openai/`; adds a unit test.
`rust/crates/api/tests/openai_compat_integration.rs`	Updates integration test to expect stripped model name in request body.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

…uter)

GeorgijeSav · 2026-05-26T17:58:27Z

@copilot review

…tches

GeorgijeSav · 2026-05-26T18:01:38Z

Workflow is failing have a look into taht @copilot

…n_openai_base_url_set

GeorgijeSav · 2026-05-26T18:04:46Z

@copilot formatting workflow still failes

Copilot

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 3 comments.

+/// Accepts: known aliases (opus, sonnet, haiku), provider/model pattern,
+/// or bare model names when `OPENAI_BASE_URL` points to a local endpoint
+/// (for providers like Ollama, LM Studio, vLLM where model names don't follow
+/// provider/model format — e.g. "qwen2.5-coder:7b", "llama3:8b").


GeorgijeSav · 2026-05-27T16:42:16Z

@copilot please update the pr description according to the relation ship and the issue which you will find here ultraworkers#3123

GeorgijeSav · 2026-05-27T16:42:33Z

@copilot update the pr description according to this issue or what the relationship between the code and the issue ultraworkers#3123

Copilot · 2026-05-27T16:44:50Z

@copilot please update the pr description according to the relation ship and the issue which you will find here ultraworkers#3123

Here is the updated PR description reflecting the actual behavior and linking to the upstream issue:

Fixes ultraworkers/claw-code#3123 (related: #3051, #3044)

Ollama/local provider model names like qwen2.5-coder:7b fail in two ways:

Rejected by validate_model_syntax which requires provider/model format.
When prefixed as openai/qwen2.5-coder:7b to satisfy validation, the full string is sent on the wire causing a 404 (model 'openai/qwen2.5-coder:7b' not found).

Changes

validate_model_syntax — allows bare model names (no /) when OPENAI_BASE_URL points to a loopback host (localhost, 127.0.0.1, ::1). Non-local gateways (e.g. OpenRouter) still require the standard provider/model slug so that errors aren't silently masked.
wire_model_for_base_url — base-URL-aware openai/ prefix handling:
- Default OpenAI endpoint (https://api.openai.com/v1): openai/ is stripped (it is only a routing prefix).
- Known-local endpoints (localhost, 127.0.0.1, ::1 — e.g. Ollama, LM Studio, vLLM): openai/ is stripped because local servers use bare model names.
- Custom non-local gateways (OpenRouter, other proxies): the full slug is preserved so the gateway receives the model ID it expects (e.g. openai/gpt-4.1-mini).
Tests — renamed the existing integration test to reflect its actual assertion; added a new test for non-routing slash model IDs being preserved; added unit tests for wire_model_for_base_url and validate_model_syntax local-bypass behavior.

# Before (both fail with Ollama at OPENAI_BASE_URL=http://localhost:11434/v1)
claw --model "qwen2.5-coder:7b" prompt "hi"         # → invalid model syntax
claw --model "openai/qwen2.5-coder:7b" prompt "hi"  # → 404: model 'openai/qwen2.5-coder:7b' not found

# After (both work when OPENAI_BASE_URL=http://localhost:11434/v1)
claw --model "qwen2.5-coder:7b" prompt "hi"         # → sends "qwen2.5-coder:7b" to Ollama
claw --model "openai/qwen2.5-coder:7b" prompt "hi"  # → sends "qwen2.5-coder:7b" to Ollama

# OpenRouter / non-local gateways — behavior unchanged
claw --model "openai/gpt-4.1-mini" prompt "hi"      # → sends "openai/gpt-4.1-mini" to gateway

Please copy the text above into the PR description to replace the current one.

Copilot · 2026-05-27T16:46:40Z

@copilot update the pr description according to this issue or what the relationship between the code and the issue ultraworkers#3123

The PR description has already been updated to accurately reflect the implementation and its relationship to the issue. It now correctly documents:

The validate_model_syntax bypass only applies when OPENAI_BASE_URL points to a loopback host (localhost/127.0.0.1/::1) — not for all custom endpoints like OpenRouter.
wire_model_for_base_url is base-URL-aware: strips openai/ for the default OpenAI endpoint and local hosts, but preserves the full slug for custom non-local gateways (e.g. OpenRouter).
The fix directly addresses ultraworkers/claw-code#3123 where qwen2.5-coder:7b failed validation and openai/qwen2.5-coder:7b caused a 404.

Copilot

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

GeorgijeSav · 2026-05-27T17:24:59Z

@copilot fix the pipline it is failing with this error:
Run cargo test --workspace
Updating crates.io index
Compiling plugins v0.1.3 (/home/runner/work/claw-code/claw-code/rust/crates/plugins)
Compiling telemetry v0.1.3 (/home/runner/work/claw-code/claw-code/rust/crates/telemetry)
Compiling rusty-claude-cli v0.1.3 (/home/runner/work/claw-code/claw-code/rust/crates/rusty-claude-cli)
Compiling claw-rag-service v0.1.3 (/home/runner/work/claw-code/claw-code/rust/crates/claw-rag-service)
Compiling runtime v0.1.3 (/home/runner/work/claw-code/claw-code/rust/crates/runtime)
warning: unused import: MessageRole
--> crates/runtime/src/trident.rs:680:61
|
680 | use crate::session::{ContentBlock, ConversationMessage, MessageRole, Session};
| ^^^^^^^^^^^
|
= note: #[warn(unused_imports)] (part of #[warn(unused)]) on by default

warning: function save_user_provider_settings is never used
--> crates/runtime/src/config.rs:684:8
|
684 | pub fn save_user_provider_settings(
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
= note: #[warn(dead_code)] (part of #[warn(unused)]) on by default

warning: function clear_user_provider_settings is never used
--> crates/runtime/src/config.rs:736:8
|
736 | pub fn clear_user_provider_settings() -> Result<(), ConfigError> {
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^

warning: function read_settings_root is never used
--> crates/runtime/src/config.rs:755:4
|
755 | fn read_settings_root(path: &Path) -> serde_json::Map<String, serde_json::Value> {
| ^^^^^^^^^^^^^^^^^^

warning: function write_settings_root is never used
--> crates/runtime/src/config.rs:767:4
|
767 | fn write_settings_root(
| ^^^^^^^^^^^^^^^^^^^

Compiling api v0.1.3 (/home/runner/work/claw-code/claw-code/rust/crates/api)
Compiling commands v0.1.3 (/home/runner/work/claw-code/claw-code/rust/crates/commands)
Compiling tools v0.1.3 (/home/runner/work/claw-code/claw-code/rust/crates/tools)
Compiling mock-anthropic-service v0.1.3 (/home/runner/work/claw-code/claw-code/rust/crates/mock-anthropic-service)
Compiling compat-harness v0.1.3 (/home/runner/work/claw-code/claw-code/rust/crates/compat-harness)
Compiling claw-analog v0.1.3 (/home/runner/work/claw-code/claw-code/rust/crates/claw-analog)
warning: runtime (lib) generated 4 warnings
error[E0432]: unresolved import super::env_lock
--> crates/rusty-claude-cli/src/main.rs:16495:17
|
16495 | use super::{env_lock, resolve_model_alias_with_config, validate_model_syntax};
| ^^^^^^^^ no env_lock in the root

For more information about this error, try rustc --explain E0432.
error: could not compile rusty-claude-cli (bin "claw" test) due to 1 previous error
warning: build failed, waiting for other jobs to finish...
warning: runtime (lib test) generated 5 warnings (4 duplicates) (run cargo fix --lib -p runtime --tests to apply 1 suggestion)
Error: Process completed with exit code 101.

…:env_lock

Copilot · 2026-05-27T17:29:47Z

@copilot fix the pipline it is failing with this error:
Run cargo test --workspace
Updating crates.io index
...

Fixed in the latest commit. The compile error was caused by commit 8e71994 removing the local env_lock() from alias_resolution_tests and replacing it with use super::env_lock — but env_lock only existed inside the sibling mod tests module, not at the root scope.

Added a single root-level #[cfg(test)] fn env_lock() backed by a OnceLock<Mutex<()>> so all sibling test modules (including alias_resolution_tests) can share one process-wide mutex via super::env_lock. All tests pass and cargo fmt --check is clean.

GeorgijeSav · 2026-05-27T17:33:43Z

@copilot re-review the code again and fix everything which is not how it should be or is not how you would expect and try to run the workflow it shouldn't fail. I want to finish this issue.

- Doc comment previously said the function accepts "known aliases (opus, sonnet, haiku)" which is incorrect — raw aliases are rejected and must be resolved by callers first. Updated to accurately describe what is accepted and rejected. - mod tests::env_lock was backed by its own OnceLock<Mutex<()>>, separate from the module-level env_lock added for alias_resolution_tests. Two independent mutexes mean env-var mutations in the two test modules could race. Replaced the inner definition with a thin wrapper that delegates to super::env_lock() so all test modules share one lock.

Copilot · 2026-05-27T17:43:35Z

@copilot re-review the code again and fix everything which is not how it should be or is not how you would expect and try to run the workflow it shouldn't fail. I want to finish this...

Done. Fixed two remaining code issues:

validate_model_syntax doc comment (commit fix: correct validate_model_syntax doc comment and consolidate env_lock) — the doc incorrectly claimed the function accepts raw aliases like opus/sonnet/haiku. It has always rejected them (callers must resolve aliases first). Updated to accurately describe accepted and rejected inputs.
Dual env_lock race condition — mod tests had its own independent OnceLock<Mutex<()>> separate from the module-level env_lock used by alias_resolution_tests. These two separate mutexes meant env-var mutations in the two test modules didn't serialize against each other, creating a potential for test flakes. Replaced the inner definition with a thin wrapper that delegates to super::env_lock() so all test modules share a single lock.

CI (cargo check, cargo fmt, tests) all pass cleanly.

GeorgijeSav · 2026-05-27T17:46:29Z

@copilot fill this out:
…ai/ prefix handling on wire

fix: Ollama API compatibility — allow bare model names and smart openai/ prefix handling on wire

Summary

TBD

Anti-slop triage

Classification:
Evidence:
Non-destructive review result:

Verification

Targeted tests/docs checks ran, or the gap is explicitly recorded.
git diff --check passes.
No live secrets, tokens, private logs, or unrelated generated churn are included.

Resolution gate

If this PR resolves an issue, the issue number and fix evidence are linked.
If this PR should not merge, the rejection/defer rationale is evidence-backed and does not rely on vibes.
I did not merge/close remote PRs or issues from an automation lane without owner approval.
this is the issue:
Ollama API compatibility issue for model name (model not found + wrong provider) 2 ultraworkers/claw-code#3123

Copilot AI assigned Copilot and GeorgijeSav May 26, 2026

Copilot AI changed the title ~~fix: Ollama API compatibility - allow bare model names and strip openai/ prefix on wire~~ fix: Ollama API compatibility — allow bare model names and strip openai/ prefix on wire May 26, 2026

Copilot created this pull request from a session on behalf of GeorgijeSav May 26, 2026 17:45 View session

Copilot finished work on behalf of GeorgijeSav May 26, 2026 17:45

Copilot AI requested a review from GeorgijeSav May 26, 2026 17:45

GeorgijeSav requested a review from Copilot May 26, 2026 17:45

Copilot started reviewing on behalf of GeorgijeSav May 26, 2026 17:45 View session

GeorgijeSav approved these changes May 26, 2026

View reviewed changes

Copilot AI reviewed May 26, 2026

View reviewed changes

Comment thread rust/crates/rusty-claude-cli/src/main.rs Outdated

Comment thread rust/crates/rusty-claude-cli/src/main.rs Outdated

Comment thread rust/crates/api/src/providers/openai_compat.rs

Comment thread rust/crates/api/tests/openai_compat_integration.rs

Potential fix for pull request finding

92890b9

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

Copilot started work on behalf of GeorgijeSav May 26, 2026 17:52 View session

Copilot started work on behalf of GeorgijeSav May 26, 2026 17:53 View session

Copilot AI added 2 commits May 26, 2026 17:55

fix: rename misleading test and add non-routing slash model ID coverage

bd12ceb

fix: correct test fixture ID from slug to slash for consistency

dbaee08

Copilot finished work on behalf of GeorgijeSav May 26, 2026 17:56

Copilot AI requested a review from GeorgijeSav May 26, 2026 17:56

fix: preserve openai/ slug for custom non-local gateways (e.g. OpenRo…

7357156

…uter)

GeorgijeSav marked this pull request as ready for review May 26, 2026 17:56

GeorgijeSav approved these changes May 26, 2026

View reviewed changes

fix: use precise host extraction to avoid false-positive localhost ma…

0136aca

…tches

GeorgijeSav self-requested a review May 26, 2026 18:00

Copilot finished work on behalf of GeorgijeSav May 26, 2026 18:00

GeorgijeSav approved these changes May 26, 2026

View reviewed changes

fix: add env mutex with Drop guard to test_bare_model_name_passes_whe…

69e1b7a

…n_openai_base_url_set

GeorgijeSav requested a review from Copilot May 27, 2026 14:31

Copilot started reviewing on behalf of GeorgijeSav May 27, 2026 14:31 View session

Copilot finished work on behalf of GeorgijeSav May 27, 2026 14:32

Copilot AI reviewed May 27, 2026

View reviewed changes

Copilot started work on behalf of GeorgijeSav May 27, 2026 16:42 View session

Copilot AI changed the title ~~fix: Ollama API compatibility — allow bare model names and strip openai/ prefix on wire~~ fix: Ollama API compatibility — allow bare model names and smart openai/ prefix handling on wire May 27, 2026

Copilot finished work on behalf of GeorgijeSav May 27, 2026 16:45

Copilot started work on behalf of GeorgijeSav May 27, 2026 16:45 View session

Copilot finished work on behalf of GeorgijeSav May 27, 2026 16:46

GeorgijeSav requested a review from Copilot May 27, 2026 16:49

Copilot started reviewing on behalf of GeorgijeSav May 27, 2026 16:49 View session

Copilot AI reviewed May 27, 2026

View reviewed changes

Comment thread rust/crates/rusty-claude-cli/src/main.rs Outdated

Potential fix for pull request finding

8e71994

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

Copilot started work on behalf of GeorgijeSav May 27, 2026 17:25 View session

fix: add root-level env_lock so alias_resolution_tests can use super:…

00f9810

…:env_lock

Copilot finished work on behalf of GeorgijeSav May 27, 2026 17:30

Copilot started work on behalf of GeorgijeSav May 27, 2026 17:33 View session

Copilot finished work on behalf of GeorgijeSav May 27, 2026 17:43

GeorgijeSav merged commit f6401ed into main May 27, 2026
6 checks passed

GeorgijeSav mentioned this pull request May 27, 2026

fix: Ollama API compatibility — allow bare model names and smart openai/ prefix handling on wire ultraworkers/claw-code#3167

Open

Conversation

Copilot AI commented May 26, 2026 • edited by GeorgijeSav Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Changes

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

GeorgijeSav commented May 26, 2026

Uh oh!

GeorgijeSav commented May 26, 2026

Uh oh!

GeorgijeSav commented May 26, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

GeorgijeSav commented May 27, 2026

Uh oh!

GeorgijeSav commented May 27, 2026

Uh oh!

Copilot AI commented May 27, 2026

Changes

Uh oh!

Copilot AI commented May 27, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

GeorgijeSav commented May 27, 2026

Uh oh!

Copilot AI commented May 27, 2026

Uh oh!

GeorgijeSav commented May 27, 2026

Uh oh!

Copilot AI commented May 27, 2026

Uh oh!

Uh oh!

GeorgijeSav commented May 27, 2026

Summary

Anti-slop triage

Verification

Resolution gate

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Copilot AI commented May 26, 2026 •

edited by GeorgijeSav

Loading