- adapter: add VertexAIAdapter with transport options and enhance response handling
- add FileCache for persisting subcall results and improve performance
- docs: update usage examples and improve context handling in README
- cache: separate cache keys by model and make FileCache crash-safe
- RLM fields
max_tokensandsubcall_max_tokensare renamed tomax_output_tokensandsubcall_max_output_tokens.
- rlm: rename max_tokens to max_output_tokens and raise defaults
- compaction: auto-resolve context window for pct threshold; deprecate max_history_tokens
- improve documentation and clarity in README, policy, prompts, and tests
- policy: update token budget handling to default to unlimited and improve exception management
- rlm: child-RLM recursion behind recursion_impl flag
- adapters: granular httpx timeouts, Retry-After, and RemoteProtocolError retry in GenericChatAdapter
- rlm: loop-collapse guard for unproductive trajectories
- rlm: conversation-history compaction with tiktoken token counting
- env: per-exec CPU-time timeout for PythonREPL
- vertex: resilient Gemini responses and gRPC deadlock guard
- add reranking and multi-query primitives + OBLIQ-Bench experiments (#32)
- rlm: improve code block extraction logic to handle trailing text and multiple fences (#31)
- examples: add demo for counting 'r' in generated fruit names and interactive trace viewer
- trace-viewer: add interactive TUI for visualizing execution traces
- azure-openai, generic-chat: update temperature parameter to be optional and improve payload handling
- azure-openai: enhance check_azure_connection to support multiple models and improve error handling
- azure-openai: handle unsupported temperature parameter and retry with default value
- tests: skip trace viewer tests if optional dependencies are not installed
- rlm: enhance prompt summary truncation and add root query support
- ci: push tag explicitly in release workflow
- doctools: Implement document access tools and policies
- extend retrieval with page-level ops, REPL extensions, diagnostics, and hardening
- add retrieval functionality to RLM with Elasticsearch support
- Update rich REPL demo for Azure GPT-5.1 and enhance environment variable handling
- Add AzureOpenAIAdapter and connectivity check for Azure environment variables
- Implemented llm_batch for parallel processing of prompts and enhance traceability
- enhance RLM to validate synchronous retriever methods and update code pattern matching
- enable TLS certificate verification for ElasticsearchRetriever and update AsyncElasticsearchRetriever documentation
- enhance search functions to support deduplication and improve prompt handling
- code review fixes for generic_chat, env, context, retrieval
- pin pydantic-monty<0.0.8 and skip tests for unimplemented functions
- updated agents.md
- pass log_truncate_code into _run_recursive_subcall to fix F821 lint errors
- multiple corrections
- Added Oolong benchmark script for baseline vs RLM evaluation
- add live rich trace functionality and event listener for RLM execution
- enhance REPL environment and prompts, add diagnostic run example
- add Vertex AI integration and testing framework
- enhanced RLM functionality and improve prompt handling
- add GitHub Actions workflow for automated release process
- add multi-turn conversation history to RLM loop (#13)
- add .claude to .gitignore
- integrate pydantic-monty as secure REPL backend (#9)
- improve adapters (#8)
- added dotenv and removed unnecessary examples
- added ai skills
- published package to pip and renaming the package in order to publish on pip (#4)
- add result export functionality and enhance README with usage details
- add demo for RLM vs baseline comparison and enhance benchmark documentation
- enhance rlm_vs_baseline benchmark for MIT RLM paper demo
- Add SmartRouter for automatic baseline/RLM selection
- enhance Context and RLM functionality with document support and recursive subcalls
- Implement core functionality for Recursive Language Model (RLM)