# algorithmicsuperintelligence/optillm

Optimizing inference proxy for LLMs

Repository: https://github.com/algorithmicsuperintelligence/optillm
Canonical: https://ross.abutalabs.com/products/optillm
Language: Python
License: Apache-2.0
License Family: permissive
Topics: agent, agentic-ai, agentic-workflow, agents, api-gateway, genai, large-language-models, llm, llm-inference, llmapi, mixture-of-experts, moa, openai, openai-api, optimization, proxy-server, agentic-framework, chain-of-thought, monte-carlo-tree-search, prompt-engineering
Last push: 2026-07-18T12:56:27+00:00

## Health v2 (maintenance only)
Score: 81/100 (v2, computed 2026-09-03T02:39:23.370411+00:00)
- activity 93, release rhythm 81, longevity 52
- inputs: {"age_days": 741, "days_push": 46, "days_rel": 46, "gap_med": 1, "n_releases_24m": 100}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 4252, forks 385 (observed 2026-08-28T04:08:40.446674+00:00)

## What it is
OptiLLM is an OpenAI API-compatible optimizing inference proxy that applies 20+ inference-time techniques (e.g., Mixture of Agents, MCTS, chain-of-thought) to improve LLM accuracy without training. It acts as a drop-in replacement server in front of OpenAI, Anthropic, Google, Cerebras, and 100+ models via LiteLLM.

## Use cases
- improve llm accuracy on math and logic problems without fine-tuning
- proxy openai api calls through reasoning optimization techniques
- get frontier-model performance from smaller cheaper models
- run best-of-n sampling or mixture of agents over any openai-compatible endpoint
- boost coding and reasoning benchmark scores at inference time
- route multiple llm providers behind one openai-compatible api

## When to choose
- you want better reasoning accuracy from existing models with zero training
- you need a drop-in OpenAI-compatible proxy for multiple providers
- you can afford extra inference-time compute to improve answer quality

## When to avoid
- you need minimal latency and cannot tolerate extra inference compute
- your workload is simple chat where optimization overhead adds no value
- you need a fully managed hosted service rather than self-hosting

## Facets
- artifact type: service
- maturity: active
- function: llm-inference, proxy, api-gateway, prompt-engineering, agent-framework, middleware
- domain: large-language-models, artificial-intelligence, apis, developer-tools
- platform: python, self-hosted, cli
- tags: openai-compatible, inference-optimization, mixture-of-agents, mcts, chain-of-thought, reasoning, test-time-compute, litellm, docker

## Member repositories
- algorithmicsuperintelligence/optillm (main) score 81

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:08:40.446674+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T18:22:10.537172+00:00, confidence not recorded.
  - readme: https://github.com/algorithmicsuperintelligence/optillm (fetched 2026-08-28T04:08:40.446674+00:00, sha 474c34d3246e)
- Data as of 2026-08-30T08:39:29.467469+00:00.