# memodb-io/Acontext

Agent Skills as a Memory Layer

Repository: https://github.com/memodb-io/Acontext
Canonical: https://ross.abutalabs.com/products/acontext
Homepage: https://acontext.io
Language: JavaScript
License: Apache-2.0
License Family: permissive
Topics: agent, context-engineering, data-platform, self-learning, agent-development-kit, ai-agent, llm, memory, self-evolving, anthropic, openai, agent-observability, context-data-platform, llm-observability, llmops
Last push: 2026-07-14T18:35:09+00:00

## Health v2 (maintenance only)
Score: 70/100 (v2, computed 2026-09-03T02:39:23.370411+00:00)
- activity 92, release rhythm 66, longevity 29
- inputs: {"age_days": 413, "days_push": 50, "days_rel": 147, "gap_med": 0.0, "n_releases_24m": 279}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 3676, forks 333 (observed 2026-08-28T04:08:13.685384+00:00)

## What it is
Acontext is an open-source memory layer for AI agents that captures learnings from agent runs and stores them as human-readable Markdown skill files. It provides SDKs (Python/TypeScript), session storage, task tracking, and a learning loop so agents can reuse distilled skills across frameworks, LLMs, and runs.

## Use cases
- give my ai agent persistent memory that learns from its runs
- store agent memory as readable markdown skill files instead of embeddings
- let my claude code agent improve itself over time
- capture lessons from agent sessions and reuse them next run
- share agent skills across different llm frameworks
- debug and inspect what my agent actually remembers
- self-host an agent memory service with session storage

## When to choose
- you want transparent, human-editable agent memory as plain Markdown files rather than opaque embeddings
- you want agents to learn procedures and warnings from task outcomes, not just store chat facts
- you need memory portable across LangGraph, Claude, AI SDK, or any file-reading framework
- you want session storage, task tracking, and skill learning in one self-hostable service

## When to avoid
- you only need semantic search over static documents - a vector store or RAG pipeline is simpler
- you need fact-based conversational memory like Mem0 or Zep rather than procedural skill learning
- you want a fully embedded library with no service dependency - Acontext runs as a platform/service

## Facets
- artifact type: service
- maturity: active
- function: agent-framework, rag, llm-inference, sdk, developer-tools, self-hosted
- domain: large-language-models, developer-tools
- platform: python, self-hosted, cloud
- tags: agent-memory, skill-files, context-engineering, llm-observability, markdown-memory, session-storage, claude-code, self-learning-agents, ai-agents, automation, nodejs, docker

## Member repositories
- memodb-io/Acontext (main) score 70

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:08:13.685384+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 2026-08-29T18:31:16.490394+00:00, confidence not recorded.
  - readme: https://github.com/memodb-io/Acontext (fetched 2026-08-28T04:08:13.685384+00:00, sha 5d1d77ffb056)
  - homepage: https://acontext.io (fetched 2026-08-29T09:25:39.450133+00:00, sha 3ab6db27ef42)
  - site_page: https://docs.acontext.io (fetched 2026-08-29T09:25:39.459656+00:00, sha 57fca1e91fa2)
  - site_page: https://docs.acontext.io/learn/quick (fetched 2026-08-29T09:25:39.461851+00:00, sha 11b7d03669ba)
- Data as of 2026-08-30T08:39:29.467469+00:00.