# lupantech/AgentFlow

AgentFlow: In-the-Flow Agentic System Optimization

Repository: https://github.com/lupantech/AgentFlow
Canonical: https://ross.abutalabs.com/products/agentflow
Homepage: https://agentflow.stanford.edu
Language: Python
License: MIT
License Family: permissive
Topics: agentic-ai, agentic-systems, llms, llms-reasoning, multi-agent-systems, reinforcement-learning, tool-augmented
Last push: 2026-02-08T14:00:53+00:00

## Health v2 (maintenance only)
Score: 47/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 66, release rhythm 35, longevity 24
- inputs: {"age_days": 341, "days_push": 206, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2017, forks 235 (observed 2026-08-28T04:06:06.146380+00:00)

## What it is
AgentFlow is a trainable, tool-integrated agentic framework that coordinates planner, executor, verifier, and generator modules through an evolving memory. It introduces Flow-GRPO, an on-policy reinforcement learning method that optimizes the planner inside multi-turn agent loops by broadcasting trajectory-level outcomes to each turn.

## Use cases
- train an llm agent with reinforcement learning in multi-turn tool-use environments
- optimize a planner for long-horizon agentic tasks with sparse rewards
- build a multi-module agent with planner, executor, verifier, and generator
- improve search, math, and scientific reasoning with tool-augmented llms
- run on-policy rl training for agentic systems instead of offline training
- evaluate a 7b agent across diverse reasoning benchmarks

## When to choose
- you want to train (not just prompt) a multi-module agentic system on-policy
- you need credit assignment across long multi-turn agent trajectories
- you're doing research on agentic rl and tool-augmented reasoning
- you want a modular planner/executor/verifier architecture with evolving memory

## When to avoid
- you need a simple training-free agent orchestration library
- you lack GPU resources for reinforcement learning training
- you need a production agent framework with broad tool integrations rather than a research codebase
- your use case is single-turn prompting without tool use

## Facets
- artifact type: framework
- maturity: active
- function: agent-framework, machine-learning, llm-training, reinforcement-learning, rag
- domain: artificial-intelligence, large-language-models, reinforcement-learning, machine-learning
- platform: python, cross-platform
- tags: multi-agent, flow-grpo, tool-use, planner-executor-verifier, on-policy-training, research, llm-reasoning, ai-agents, gpu

## Member repositories
- lupantech/AgentFlow (main) score 47

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:06.146380+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 2026-08-30T03:00:40.967283+00:00, confidence not recorded.
  - readme: https://github.com/lupantech/AgentFlow (fetched 2026-08-28T04:06:06.146380+00:00, sha d59ee9dc30a4)
  - homepage: https://agentflow.stanford.edu (fetched 2026-08-29T10:40:33.133630+00:00, sha 41a93d9b85d4)
- Data as of 2026-08-30T08:39:29.467469+00:00.