>_
.issue.db
/highway
Dashboard
Issues
Memory
Lessons
Audit Log
New Issue
Edit Issue #754
Update issue details
Title *
Description
Empirical (demo tenant): functiongemma:270m emits malformed/empty tool args on compound prompts (run_shell arguments={}), so the agent fails multi-step tasks and reports it could not execute. gemma4:31b-cloud DOES function-call well (emitted 'ls /tmp | wc -l && date'); deepseek-v3.1:671b-cloud and qwen3-coder:480b-cloud emit proper parallel tool calls; gpt-oss:120b-cloud works; qwen3-vl:235b-instruct is RETIRED (HTTP 410). Validated end-to-end (run 019ee788): function_model=gemma4:31b-cloud (+ function_base_url=https://ollama.com) completes the compound task correctly. RECOMMEND: collapse to a single capable model (gemma4:31b-cloud for both tool-decision and answer) - the two-model split assumed gemma4 cannot function-call, which is false. Update run_goal defaults + agent_run_goal.py + MCP embedded DSL. Trade-off: cloud cost/latency vs local 270M; user decision.
Priority
Low
Medium
High
Critical
Status
Open
In Progress
Closed
Won't Do
Due Date (YYYY-MM-DD)
Tags (comma separated)
Related Issues (IDs)
Enter IDs of issues related to this one. They will be linked as 'related'.
Update Issue
Cancel