0.1.9

✅ Key New Features

• Added chunk-processing pipeline under content_processing.chunk_processing with map_chunks and map_paragraphs modes (provider: semchunks), including overlap and output limits.
• Added preprocessing-only execution behavior for empty agent bodies (chain, instructions, and system_prompt all empty), returning sampled/chunked content without an LLM call.
• Added execution logging entries with call-site context and cURL reconstruction support for request debugging.
• Added tool-call dependency memory (state + memory) and model profile/tool-mode options to support smaller tool-calling model setups.

🛠️ Improvements & Stabilization

• Renamed model setting max_tokens to max_completion_tokens across runtime/spec usage.
• Added orchestrator agent-state retention (orchestrator.agent) and optional injected OpenAI client pass-through.
• Improved agent loading validation with clearer missing-agent errors and YAML ParserError handling.
• Added CLI validation for missing user agents directory before run.
• Expanded output handling to support list outputs (including chunked result arrays) and handoff return propagation.

🧪 Testing & Quality

• Added dedicated chunking test suite: src/tests/test_map_chunking.py.
• Expanded orchestrator tests for chunk/sample execution paths, execution logging behavior, memory deps, and handoff return behavior.
• Updated HTTP/tooling tests and integration tests for new model parameter naming and runtime behavior.

📚 Documentation & Templates

• Added chunking documentation: docs/chunking.md.
• Added large-context examples under examples/large_context/ (map_chunks, map_paragraphs, sample_then_chunk, and summary agent examples).
• Updated docs/advanced.md with preprocessing-only run behavior and sample/chunk-only examples.
• Updated README.md docs index to include chunking docs and refreshed template/example references.

Full Changelog: 0.1.8...0.1.9

0.1.8

✅ Key New Features

• Large-input sampling added via content_processing.sample (head/middle/footer ratios + max token budget).
• Sample debug output file support added to write sampled content for inspection.
• New sample repo example added at examples/sample_25_50_25/ (agent config, 6000+ token input, usage README).

🛠️ Improvements & Stabilization

• Added advanced model request controls: extra_headers and extra_body in model/runtime settings.
• Added runtime metrics capture for model calls (usage/model/response metadata).
• Updated runtime flow to apply sampling before normal chain execution, with supporting model/spec wiring.

🧪 Testing & Quality

• Added dedicated sampler test suite (test_sample_ratios.py) and expanded orchestrator coverage.
• Extended tests for HTTP/model settings and metrics behavior.
• CI workflow updated (.github/workflows/tests.yml).

📚 Documentation & Templates

• Added advanced docs (docs/advanced.md) for extra_headers/extra_body.
• Added sample feature docs (docs/sample.md) with configuration and run instructions.
• Updated README.md to include links to sample docs and advanced docs.

Full Changelog: 0.1.7...0.1.8

0.1.7

✅ Key New Features

• Single-shot execution mode added (runs one prompt/response cycle without tool orchestration).
• File manager tooling introduced (filesystem tools + adapter support).
• More flexible output handling (new output options/formats and better structured outputs).

🛠️ Improvements & Stabilization

• Switched to pydantic-ai-slim to reduce dependency weight.
• Added default timeouts and overall cleanup/refactoring for safer behavior.
• Improved type safety (type cleanups + fixes across the codebase).

🧪 Testing & Quality

• Expanded unit/integration tests and added new fixtures.
• Updated docs to match new behavior (single-shot, templates, outputs, CLI docs).

📚 Documentation & Templates

• Added/expanded docs for:
  • Single shot mode
  • Output formats
  • Agent templates (now include Instructions + System Prompt sections)

0.1.2

• Added input adaptors (FileInput, TextInput) with CLI/orchestrator wiring and tests.
• Refined chain/state handling in QuickAgent, expanded tests, and documented state usage.
• Added/updated docs and examples (including llms.txt, new business extraction agents).