v1.1.0

What's Changed

Features

• Structured outputs — response_format support threaded through LM, orchestrator, and algorithms (#211)
• Raw choice preservation — include_raw_choices option to retain full API choice metadata on responses (#226, #227)
• LLM Judge improvements — robust JSON extraction (markdown fences, truncated JSON, regex fallback), structured output schema for scoring
• Coding agent plugin — .claude-plugin/ and .codex-plugin/ with skills for inference scaling, batch scaling, setup guide
• Algorithm animations — ManimGL visualizations for self-consistency, best-of-n, particle filtering (#225)
• E2e test suite — async and sync mode end-to-end tests with AIME/MATH500 subsets (#212)

Fixes

• TaskGroup error handling — wraps ExceptionGroup into a descriptive RuntimeError with counts and type summary (#233)
• Retryable errors expanded — RETRYABLE_ERRORS now includes ConnectionError, TimeoutError, OSError, and aiohttp.ClientError
• Concurrency race in test mocks — StepMockLanguageModel fix (#221)
• Circular reference in include_raw_choices (#227)
• Eval harness — removed phantom type_check dimension, redistributed weights (#224)

Chores / CI

• Better test coverage workflows (#219, #234)
• Dependabot + Mergify configs, issue/PR templates
• Removed factory.md config file (#238)
• Docs sidebar navigation fixes (#222)
• Default max_tries reduced from 8 to 3

API Changes

• AbstractOrchestrator.agenerate() signature simplified: explicit max_tokens, temperature, include_stop_str_in_output, tools, tool_choice params replaced with **kwargs. Callers using keyword args are unaffected.

Full Changelog: v1.0.0...v1.1.0

v1.0.0

What's Changed

• Rename selection_method to final_response_selection for clarity by @gx-ai-architect in #175
• Restructure for better api and core implementation separation by @s-akhtar-baig in #186
• Add orchestration pattern to provide structured concurrency for LM calls by @s-akhtar-baig in #190
• Clean up session for current event loop in algorithm::infer by @s-akhtar-baig in #204
• Add support for newer vLLM versions in reward_hub by @s-akhtar-baig in #206
• Update docs to reflect refactor and orchestrator changes by @s-akhtar-baig in #205

New Contributors

• @s-akhtar-baig made their first contribution in #186

Full Changelog: v0.3.5...v1.0.0

v0.3.5 - Entropic Particle Filtering

What's Changed

• update documentation with its-hub latest install / usage guides by @gx-ai-architect in #173
• feat/entropic_particle_filtering by @georgosgeorgos in #144

New Contributors

• @georgosgeorgos made their first contribution in #144

Full Changelog: v0.3.4...v0.3.5

v0.3.4 - Dedup response. Update workflow. Add back uvicorn dependency

What's Changed

• dedup responses before sending to reward models for scoring by @gx-ai-architect in #164
• spec: ITS Gateway Production Readiness with Envoy External Processor by @beatsmonster in #166
• Update python build action for 3.14 compatibility by @Maxusmusti in #170
• Add uvicorn to project dependencies by @gx-ai-architect in #172

New Contributors

• @beatsmonster made their first contribution in #166

Full Changelog: v0.3.3...v0.3.4

v0.3.3 upgrade reward-hub; remove uvicorn bound

What's Changed

• Update reward-hub dependency to version 0.1.7 by @gx-ai-architect in #162

Full Changelog: v0.3.2...v0.3.3

v0.3.2 fix-bon processing bug; upgrade dependency range;

What's Changed

• Support multi modal input by @gx-ai-architect in #159
• bug fix: stringify tool-call by @gx-ai-architect @mihirathale98 in #160
• relax dependency bound by @gx-ai-architect in #161

Full Changelog: v0.3.1...v0.3.2

v0.3.1 Release include Litellm integration, llm-judge reward, reward-hub upgrade

What's Changed

• Feat/async support by @mihirathale98 in #127
• fix for budget 1 by @mihirathale98 in #128
• Fix/self consistency filtering by @gx-ai-architect in #131
• Feat/llm judge integration by @mihirathale98 in #151
• Feat/litellm integration by @mihirathale98 in #152
• add hierarchical tool-calling notebook by @gx-ai-architect in #156
• fixed judge endpoint validation and updated justfile by @mihirathale98 in #155
• updated dep reward hub version by @mihirathale98 in #157

Full Changelog: v0.3.0...v0.3.1

v0.3.0

What's Changed

• Lower Python version requirement to >=3.10 by @xukai92 in #153

Full Changelog: v0.2.5...v0.3.0

v0.2.5

What's Changed

• Feat/async implementation by @mihirathale98 in #140
• Lower pydantic version requirement to >=2.0.0 by @xukai92 in #150

Full Changelog: v0.2.4...v0.2.5

Release endpoint feature & upgrade vllm dependency version

What's Changed

• feat: Add ParticleFilteringResult with flattened structure by @shivchander in #103
• exclude request args for openai endpoints by @gx-ai-architect in #108
• Support hierarchical consistency space and regex projection functions by @xukai92 in #109
• Implement ChatMessages class to unify input handling and eliminate boilerplate by @gx-ai-architect in #119
• Fix linting issues: format code and remove trailing whitespace by @gx-ai-architect in #123
• Implement tool calls support and LM response format updates by @gx-ai-architect in #121
• fix for hashing tool args by @mihirathale98 in #124
• Implement tool-vote feature for self-consistency algorithm by @gx-ai-architect in #122
• Bug/fix bias for tool by @gx-ai-architect in #126
• fix for budget 1 by @gx-ai-architect in #130
• Self consistency fix by @gx-ai-architect in #139
• Feature/notebook update by @gx-ai-architect in #141
• Update pyproject.toml by @gx-ai-architect in #142

New Contributors

• @shivchander made their first contribution in #103

Full Changelog: v0.2.3a1...v0.2.4