ACE is an open-source Python engine that adds a persistent learning loop to AI agents, enabling them to learn from experience and improve over time. It offers multiple runtimes, a Skillbook concept, and integrations via LiteLLM and various runners.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
ACE is the open-source engine behind Kayba that adds a persistent learning loop to agents. It maintains a Skillbook of strategies and uses a Recursive Reflector to extract actionable insights from traces. The project is written in Python and is available under the Apache-2.0 license.
How it works
ACE maintains a Skillbook — a persistent collection of strategies that evolves with every task. Roles include Agent, Reflector, and SkillManager. The Recursive Reflector writes and executes Python code in a sandboxed environment to search for patterns and derive insights. The architecture relies on PydanticAI agents with structured output validation and supports multiple providers through LiteLLM integration.
Getting started
Quick Start steps are shown in the README:
uv add ace-framework
Option A — Interactive setup (recommended):
ace setup # Walks you through model selection, API keys, and connection validation
Option B — Manual configuration:
export OPENAI_API_KEY="your-key" # or ANTHROPIC_API_KEY, or any of 100+ supported providers
Then use it:
from ace import ACELiteLLM
agent = ACELiteLLM(model="gpt-4o-mini")
# First attempt — the agent may hallucinate
answer = agent.ask("Is there a seahorse emoji?")
# Feed a correction — ACE extracts a strategy and updates the Skillbook
agent.learn_from_feedback("There is no seahorse emoji in Unicode.")
# Subsequent calls benefit from the learned strategy
answer = agent.ask("Is there a seahorse emoji?")
# Inspect what the agent has learned
print(agent.get_strategies())
No fine-tuning, no training data, no vector database.
Links in the README point to Quick Start, Setup guides, and Hosted API documentation for further details.
Recent releases
- v0.12.0 includes a rewrite and hardening of Skillbook/architecture lines; merger of 0.11.0 architectural rewrite and 0.12.0 SkillManager hardening. (2026-05-07)
- openclaw-tracing-v0.1.1 — Initial release of the OpenClaw tracing plugin (2026-04-27)
- kayba-tracing-ts-v0.10.0 — TypeScript SDK release (2026-04-25)
- v0.10.0 — Added session metadata hooks and usage metering
- v0.9.7 — TypeScript SDK (2026-04-11)
Traction
2547 stars, 302 forks, 12 open issues as of the data provided.
Behind the repo
This repository is part of Kayba, with hosted services available at kayba.ai; the README notes an optional hosted API and integration guides.
Caveats
License: Apache-2.0. Created 2025-10-15; last push 2026-07-08. Language: Python. Open issues: 12. Tags indicate agent-learning, agent-memory, llm, and memory-related projects.