# ByteDance-Seed/m3-agent

Repository: https://github.com/ByteDance-Seed/m3-agent
Canonical: https://ross.abutalabs.com/products/m3-agent
Language: Python
License: Apache-2.0
License Family: permissive
Last push: 2026-02-12T06:03:56+00:00

## Health v2 (maintenance only)
Score: 48/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 67, release rhythm 35, longevity 28
- inputs: {"age_days": 399, "days_push": 202, "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 1445, forks 117 (observed 2026-08-28T04:04:45.089287+00:00)

## What it is
M3-Agent is a multimodal agent framework from ByteDance Seed that processes real-time visual and auditory inputs to build entity-centric long-term memory (episodic and semantic) and performs iterative memory-based reasoning. It is released alongside M3-Bench, a long-video question-answering benchmark, and RL-trained agent models on Hugging Face.

## Use cases
- build a multimodal agent with long-term memory
- question answering over long videos
- research on memory-based reasoning in agents
- evaluate multimodal agents on long-video benchmarks
- train an agent with reinforcement learning for video understanding
- process real-time camera and audio streams into semantic memory

## When to choose
- you need a research-grade multimodal agent with persistent, entity-centric memory
- you want to benchmark long-video QA and memory-based reasoning
- you want to build on or fine-tune the released M3-Agent models

## When to avoid
- you need a production-ready personal assistant out of the box
- you only need simple video transcription or captioning without memory
- you cannot run GPU-heavy multimodal inference

## Facets
- artifact type: framework
- maturity: active
- function: agent-framework, machine-learning, llm-training, rag, video-processing, speech-recognition, computer-vision
- domain: artificial-intelligence, large-language-models, machine-learning
- platform: python
- tags: multimodal-agent, long-term-memory, reinforcement-learning, video-qa, benchmark, iclr-2026, semantic-memory, episodic-memory, ai-agents, video, research, linux, gpu

## Member repositories
- ByteDance-Seed/m3-agent (main) score 48

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:45.089287+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 2026-08-30T04:36:12.304998+00:00, confidence not recorded.
  - readme: https://github.com/ByteDance-Seed/m3-agent (fetched 2026-08-28T04:04:45.089287+00:00, sha bc5f630539c2)
- Data as of 2026-08-30T08:39:29.467469+00:00.