Open source AI agent self-improvement
Run Emend yourself to give AI agents local memory, online playbook learning, and retrieval-backed behavior improvement without retraining the model.
pip install emend-ai
OSSemend local loop
Publish interactions
Send each user turn, agent response, correction, and outcome to the local Emend server.
Extract memory and playbooks
Emend turns durable user facts into profiles and behavioral lessons into user playbooks.
Retrieve learned context
Your agent searches profiles and playbooks before the next response, so feedback changes future behavior.
Agent self-improvement you can inspect, run, and adapt
Most agents repeat mistakes because corrections disappear after the current conversation. Emend keeps the learning loop outside the model: interactions become profiles, playbooks, and searchable context your agent can use next time.
Standalone self-hosting
Run the open-source package on your own machine or infrastructure with a local FastAPI backend.
Online learning loop
Keep publishing interactions and Emend learns from real corrections as they arrive.
Local storage by default
Start with local SQLite for development, then choose stronger storage when your deployment needs it.
Open implementation
Inspect the memory, extraction, and retrieval path instead of treating agent learning as a black box.
Start standalone. Upgrade when operations matter.
The open-source package is built for self-hosted online learning. Emend Enterprise adds managed infrastructure, stronger storage modes, governance surfaces, and continuous offline optimization for playbooks.
Open source
Enterprise
A self-improving agent loop in a few calls
Install the open-source package, run the local backend, publish interactions, then retrieve learned context before future agent turns.
from emend import InteractionData, EmendClient client = EmendClient(url_endpoint="http://localhost:8081") client.publish_interaction( user_id="user_123", interactions=[ InteractionData(role="User", content="I travel weekly."), InteractionData(role="Agent", content="Try this bulky laptop."), InteractionData( role="User", content="Too heavy. Prioritize battery life and weight.", ), ], source="local-agent", session_id="session_001",) context = client.search( query="laptop recommendation for user_123", user_id="user_123",)Open-source agent improvement, without the hand-waving
Open-source AI agent self-improvement is a feedback loop where an agent records real interactions, extracts useful memory and behavioral guidance, and retrieves that context before future responses. Emend provides this loop as an inspectable open-source package.
Build the local loop first.
If you want inspectable AI agent self-improvement today, start with the open-source quickstart. If you need managed storage, dashboards, or offline playbook optimization, explore Emend Enterprise.