Acontext

Acontext presents itself as a context data platform for building cloud-native AI agents. The README frames four problems: LLM messages make up most of an agent database and poor schema design makes them expensive and slow, long-running agents need context engineering written from scratch, you cannot see how an agent is doing, and agents are hit or miss and do not learn from their successes. Acontext's answer is to store, observe and learn.

Its core features are context engineering (Session for unified multi-provider message storage including images, audio and files, Disk for saving and downloading artifacts by file path, and Context Editing to manage the context window through one API), observation of agent tasks and user feedback with near real-time status, an experience agent that learns standard operating procedures for each user from successful runs, and a dashboard showing agent success rate and other metrics. Storage uses Postgres, Redis and S3. There are SDKs (for example `pip install acontext`) and you can self-host with `acontext docker up`, which needs Docker and an OpenAI API key and creates .env and config.yaml files and a db folder. GitHub shows no licence for the repository.