Acontext is a context data platform designed to enhance the reliability and task success rates of self-learning AI agents. It addresses the complexities of context engineering by providing a unified system for agents to store, observe, and learn from their experiences. This platform benefits developers building scalable agent products by improving agent stability and delivering greater user value through continuous learning and adaptation.

How It Works

Acontext operates around core concepts: Sessions for conversation threads, Disks for artifact storage, Task Agents for observing agent progress, Experience Agents for distilling skills, and Spaces for Notion-like storage of learned experiences (SOPs). Agent interactions are stored in Sessions, tasks are observed, and successful patterns are distilled into structured skills within a Space. These learned skills can then be searched and utilized by future agent sessions, enabling a self-learning loop that improves agent performance over time.

Quick Start & Requirements

• Primary install: CLI via curl -fsSL https://install.acontext.io | sh. Python (pip install acontext) and Typescript (npm i @acontext/acontext) SDKs available.
• Prerequisites: Docker, OpenAI API Key (recommended GPT models).
• Setup: Run acontext docker up in a dedicated directory for backend. Local API at http://localhost:8029/api/v1, Dashboard at http://localhost:3000/.
• Links: Example repository: Acontext-Examples. Extensive documentation available.

Highlighted Details

• Supports multi-modal message storage and retrieval.
• Automatically observes agent tasks, extracting progress and user preferences.
• "Space" feature provides a structured, Notion-like system for storing and searching learned agent skills (SOPs).
• Offers fast (embedding-based) and agentic (exploratory) modes for skill search.
• Includes a local Dashboard for monitoring metrics, sessions, tasks, artifacts, and learned spaces.

Maintenance & Community

The project encourages community engagement via Discord and provides updates via X. Users can star the project on GitHub for notifications. A roadmap and contribution guidelines are available in respective files (roadmap.md, contributing.md).

Licensing & Compatibility

This project is licensed under the Apache License 2.0. This license is permissive and generally compatible with commercial use and linking within closed-source applications.

Limitations & Caveats

The self-learning process involves a background distillation delay of approximately 10-30 seconds. The platform relies on OpenAI models for its core functionality.