Skip to content

Latest commit

 

History

History
129 lines (86 loc) · 4.45 KB

File metadata and controls

129 lines (86 loc) · 4.45 KB

v1.0.3 (2026-06-09)

Feat

  • adapter: add VertexAIAdapter with transport options and enhance response handling

v1.0.2 (2026-06-08)

Feat

  • add FileCache for persisting subcall results and improve performance
  • docs: update usage examples and improve context handling in README

Fix

  • cache: separate cache keys by model and make FileCache crash-safe

v1.0.1 (2026-06-05)

BREAKING CHANGE

  • RLM fields max_tokens and subcall_max_tokens are renamed to max_output_tokens and subcall_max_output_tokens.

Feat

  • rlm: rename max_tokens to max_output_tokens and raise defaults
  • compaction: auto-resolve context window for pct threshold; deprecate max_history_tokens

Refactor

  • improve documentation and clarity in README, policy, prompts, and tests

v1.0.0 (2026-06-04)

Feat

  • 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)

v0.4.2 (2026-05-14)

Fix

  • rlm: improve code block extraction logic to handle trailing text and multiple fences (#31)

v0.4.1 (2026-05-05)

Feat

  • examples: add demo for counting 'r' in generated fruit names and interactive trace viewer
  • trace-viewer: add interactive TUI for visualizing execution traces

Fix

  • 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

v0.4.0 (2026-04-08)

Feat

  • 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

Fix

  • 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

Perf

  • Added Oolong benchmark script for baseline vs RLM evaluation

v0.3.0 (2026-03-09)

Feat

  • 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

v0.2.0 (2026-02-15)

Feat

  • integrate pydantic-monty as secure REPL backend (#9)
  • improve adapters (#8)
  • added dotenv and removed unnecessary examples
  • added ai skills

Refactor

  • published package to pip and renaming the package in order to publish on pip (#4)

v0.1.0 (2026-01-21)

Feat

  • 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)