feat(agent): no-tools-called retry for local Ollama models (complements #210)

[mike-011]

Summary

Complements PR #210 — adds a no-tools-called retry for local Ollama models (12-14B) that return a final answer without calling any tools.

Problem

After PR #210 fixed the ollama/ → ollama_chat/ rewrite and ModelBehaviorError retry, testing with local models revealed a second failure mode:

• qwen2.5-coder:14b (local): add works (227s), but query returns "I cannot find relevant information" — the model didn't call read_file
• gemma4:12b (local): query returns empty output ""

The KB is correctly built (wiki, concepts, entities all present), but the model doesn't initiate tool calls during query — it answers directly without reading wiki files.

This is a model capability limitation, not a transport bug. The adapter cannot force a model to call tools, but it can nudge it.

Fix

After Runner.run() completes successfully, the adapter checks result.new_items for any ToolCallItem. If the agent has tools but none were called, the adapter retries with an escalating nudge message:

Attempt 1 (gentle):

You answered without reading any wiki files. You have tools available. Call read_file with path 'summaries/index.md' first to see what documents exist, then answer the question.

Attempt 2 (firm):

You MUST use the read_file tool. Do NOT answer without reading files first. Start with read_file(path='summaries/index.md'). Available tools: read_file, get_page_content.

Attempt 3 (final):

IMPORTANT: Your previous answer was not grounded in the wiki. Call read_file now. The available tools are: read_file, get_page_content. Do not answer until you have called a tool.

If all retries are exhausted, the adapter returns the last (ungrounded) answer — no infinite loops.

Interaction with PR #210

This PR includes all changes from PR #210 (ollama_chat/ rewrite, ModelBehaviorError retry, parameterized timeouts, drop_params, api_base propagation) plus the no-tools-called retry. The two retry mechanisms share the same tool_call_retries budget.

Testing

• 73 adapter unit tests (17 new for no-tools retry) — all passing
• 1148 total tests — all passing (0 regressions)

python -m pytest tests/test_ollama_adapter.py -v  # 73 passed
python -m pytest tests/ -q                          # 1148 passed

Files Changed

File	Description
openkb/agent/ollama_adapter.py	Added _has_tool_calls(), _build_nudge_message(), _append_nudge(), no-tools retry loop
tests/test_ollama_adapter.py	17 new tests (73 total)
docs/ollama_no_tools_retry.md	New — full documentation
(+ all files from PR #210)	ollama_chat/ rewrite, compiler, cli, config, docs

Configuration

Same ollama: config block as PR #210:

ollama:
  timeout: 300              # per-request timeout (s), default 300
  tool_call_retries: 3      # max retries (hallucination + no-tools), default 3

Limitations

• The nudge cannot force a model to call tools — it can only encourage. Very small models (1B) may ignore it.
• Models that produce empty output ("") may not benefit if the empty output is caused by a timeout or inference failure.
• The retry budget is shared with ModelBehaviorError retries.