You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Tool usage failures with Claude Sonnet 4 via openai compatible provider despite working fine with direct Claude subscription #210
V.0.1.98 Issue: Tool usage failures with Claude Sonnet 4 via HelixMind provider despite working fine with direct Claude subscription
Description
Experiencing tool usage failures when using claude-sonnet-4-20250514 through the HelixMind provider (OpenAI-compatible API), while the same model works perfectly with a direct Claude subscription.
Expected Behavior
Tool usage should work consistently regardless of whether accessing Claude through direct subscription or via HelixMind provider.
changed the title [-]Can't manage to properly use tools[/-][+]Tool usage failures with Claude Sonnet 4 via openai compatible provider despite working fine with direct Claude subscription[/+]on Jun 19, 2025
roo code does not make use of native function calling. Roo code asks the llm to mock fucntion calls by formatting the response in xml tags , and then parses the xml to identify the function associated .
It is native function callling that is broken in helixmind . In certain cases native function calling works in helixmind.
I could not yet identify when it will work and when it gets broken.
Activity
[-]Can't manage to properly use tools[/-][+]Tool usage failures with Claude Sonnet 4 via openai compatible provider despite working fine with direct Claude subscription[/+]rkunnamp commentedon Jun 19, 2025
Do not think this is specific to opencode (not sure though). Have observed similar issues with helixmind apis in other frameworks as well.
ncls-p commentedon Jun 19, 2025
Maybe it does not works the same way, i have not checked but i have no issue when using roo code with helixmind
rkunnamp commentedon Jun 19, 2025
roo code does not make use of native function calling. Roo code asks the llm to mock fucntion calls by formatting the response in xml tags , and then parses the xml to identify the function associated .
It is native function callling that is broken in helixmind . In certain cases native function calling works in helixmind.
I could not yet identify when it will work and when it gets broken.
ncls-p commentedon Jun 19, 2025
Okay, i did not know thanks for those informations !
rekram1-node commentedon Dec 28, 2025
[automated] Closing due to 90+ days of inactivity. Feel free to reopen if you still need this!
Merge pull request anomalyco#210 from Warshoow/docs/token-optimizatio…
test: fix test fixture config injection and remove dead tests (anomal…
test: fix test fixture config injection and remove dead tests (anomal…