Acontext is an open-source skill memory layer for AI agents. It automatically captures learnings from agent runs and stores them as Markdown files — files you can read, edit, and share across agents, LLMs, and frameworks.
If you want the agent you build to learn from its mistakes and reuse what worked — without opaque memory polluting your context — give Acontext a try.
Agent memory is getting increasingly complicated🤢 — hard to understand, hard to debug, and hard for users to inspect or correct. Acontext takes a different approach: if agent skills can represent every piece of knowledge an agent needs as simple files, so can the memory.
- Acontext builds memory in the agent skills format, so everyone can see and understand what the memory actually contains.
- Skill is Memory, Memory is Skill. Whether a skill comes from one you downloaded from Clawhub or one you created yourself, Acontext can follow it and evolve it over time.
- Plain file, any framework — Skill memories are Markdown files. Use them with LangGraph, Claude, AI SDK, or anything that reads files. No embeddings, no API lock-in. Git, grep, and mount to the sandbox.
- You design the structure — Attach more skills to define the schema, naming, and file layout of the memory. For example: one file per contact, one per project by uploading a working context skill.
- Progressive disclosure, not search — The agent can use
get_skillandget_skill_fileto fetch what it needs. Retrieval is by tool use and reasoning, not semantic top-k. - Download as ZIP, reuse anywhere — Export skill files as ZIP. Run locally, in another agent, or with another LLM. No vendor lock-in; no re-embedding or migration step.
flowchart LR
A[Session messages] --> C[Task complete/failed]
C --> D[Distillation]
D --> E[Skill Agent]
E --> F[Update Skills]
- Session messages — Conversation (and optionally tool calls, artifacts) is the raw input. Tasks are extracted from the message stream automatically (or inferred from explicit outcome reporting).
- Task complete or failed — When a task is marked done or failed (e.g. by agent report or automatic detection), that outcome is the trigger for learning.
- Distillation — An LLM pass infers from the conversation and execution trace what worked, what failed, and user preferences.
- Skill Agent — Decides where to store (existing skill or new) and writes according to your
SKILL.mdschema. - Update Skills — Skills are updated. You define the structure in
SKILL.md; the system does extraction, routing, and writing.
flowchart LR
E[Any Agent] --> F[list_skills/get_skill]
F --> G[Appear in context]
Give your agent Skill Content Tools (get_skill, get_skill_file). The agent decides what it needs, calls the tools, and gets the skill content. No embedding search — progressive disclosure, agent in the loop.
- Go to Acontext.io, claim your free credits.
- Go through a one-click onboarding to get your API Key (starts with
sk-ac)
💻 Self-host Acontext
We have an acontext-cli to help you do a quick proof-of-concept. Download it first in your terminal:
curl -fsSL https://install.acontext.io | shYou should have docker installed and an OpenAI API Key to start an Acontext backend on your computer:
mkdir acontext_server && cd acontext_server
acontext server upMake sure your LLM has the ability to call tools. By default, Acontext will use
gpt-4.1.
acontext server up will create/use .env and config.yaml for Acontext, and create a db folder to persist data.
Once it's done, you can access the following endpoints:
- Acontext API Base URL: http://localhost:8029/api/v1
- Acontext Dashboard: http://localhost:3000/
We're maintaining Python 
Click the doc link to see TS SDK Quickstart.
pip install acontextimport os
from acontext import AcontextClient
# For cloud:
client = AcontextClient(
api_key=os.getenv("ACONTEXT_API_KEY"),
)
# For self-hosted:
client = AcontextClient(
base_url="http://localhost:8029/api/v1",
api_key="sk-ac-your-root-api-bearer-token",
)Create a learning space, attach a session, and let the agent learn — skills are written as Markdown files automatically.
from acontext import AcontextClient
client = AcontextClient(api_key="sk-ac-...")
# Create a learning space and attach a session
space = client.learning_spaces.create()
session = client.sessions.create()
client.learning_spaces.learn(space.id, session_id=session.id)
# Run your agent, store messages — when tasks complete, learning runs automatically
client.sessions.store_message(session.id, blob={"role": "user", "content": "My name is Gus"})
client.sessions.store_message(session.id, blob={"role": "assistant", "content": "Hi Gus! How can I help you today?"})
# ... agent runs ...
# List learned skills (Markdown files)
client.learning_spaces.wait_for_learning(space.id, session_id=session.id)
skills = client.learning_spaces.list_skills(space.id)
# Download all skill files to a local directory
for skill in skills:
client.skills.download(skill_id=skill.id, path=f"./skills/{skill.name}")
wait_for_learningis a blocking helper for demo purposes. In production, task extraction and learning run in the background automatically — your agent never waits.
- Context Engineering — Compress context with summaries and edit strategies
- Disk — Virtual, persistent filesystem for agents
- Sandbox — Isolated code execution with bash, Python, and mountable skills
- Agent Tools — Disk tools, sandbox tools, and skill tools for LLM function calling
Download end-to-end scripts with acontext:
Python
acontext create my-proj --template-path "python/openai-basic"More examples on Python:
python/openai-agent-basic: openai agent sdk templatepython/openai-agent-artifacts: agent can edit and download artifactspython/claude-agent-sdk: claude agent sdk withClaudeAgentStoragepython/agno-basic: agno framework templatepython/smolagents-basic: smolagents (huggingface) templatepython/interactive-agent-skill: interactive sandbox with mountable agent skills
Typescript
acontext create my-proj --template-path "typescript/openai-basic"More examples on Typescript:
typescript/vercel-ai-basic: agent in @vercel/ai-sdktypescript/claude-agent-sdk: claude agent sdk withClaudeAgentStoragetypescript/interactive-agent-skill: interactive sandbox with mountable agent skills
Note
Check our example repo for more templates: Acontext-Examples.
We're cooking more full-stack Agent Applications! Tell us what you want!
To learn more about skill memory and what Acontext can do, visit our docs or start with What is Skill Memory?
Star Acontext on GitHub to support us and receive instant notifications.
click to open
graph TB
subgraph "Client Layer"
PY["pip install acontext"]
TS["npm i @acontext/acontext"]
end
subgraph "Acontext Backend"
subgraph " "
API["API<br/>localhost:8029"]
CORE["Core"]
API -->|FastAPI & MQ| CORE
end
subgraph " "
Infrastructure["Infrastructures"]
PG["PostgreSQL"]
S3["S3"]
REDIS["Redis"]
MQ["RabbitMQ"]
end
end
subgraph "Dashboard"
UI["Web Dashboard<br/>localhost:3000"]
end
PY -->|RESTFUL API| API
TS -->|RESTFUL API| API
UI -->|RESTFUL API| API
API --> Infrastructure
CORE --> Infrastructure
Infrastructure --> PG
Infrastructure --> S3
Infrastructure --> REDIS
Infrastructure --> MQ
style PY fill:#3776ab,stroke:#fff,stroke-width:2px,color:#fff
style TS fill:#3178c6,stroke:#fff,stroke-width:2px,color:#fff
style API fill:#00add8,stroke:#fff,stroke-width:2px,color:#fff
style CORE fill:#ffd43b,stroke:#333,stroke-width:2px,color:#333
style UI fill:#000,stroke:#fff,stroke-width:2px,color:#fff
style PG fill:#336791,stroke:#fff,stroke-width:2px,color:#fff
style S3 fill:#ff9900,stroke:#fff,stroke-width:2px,color:#fff
style REDIS fill:#dc382d,stroke:#fff,stroke-width:2px,color:#fff
style MQ fill:#ff6600,stroke:#fff,stroke-width:2px,color:#fff
Join the community for support and discussions:
- Check our roadmap.md first.
- Read contributing.md
[](https://acontext.io)
[](https://acontext.io)This project is currently licensed under Apache License 2.0.
![@dependabot[bot]](https://avatars.githubusercontent.com/in/29110?s=64&v=4){kind=small}
