agentic-context-engine

🧠 Make your agents learn from experience. Now available as a hosted solution at kayba.ai

What is agentic-context-engine?

🧠 Make your agents learn from experience. Now available as a hosted solution at kayba.ai

Overview of agentic-context-engine

What it is

Agentic Context Engine (ACE) is an open-source engine that adds a persistent learning loop to AI agents. It is the engine behind Kayba, a managed service. ACE maintains a Skillbook—a collection of strategies that evolves with each task. Three specialized roles manage the learning loop: Agent (executes tasks with Skillbook strategies), Reflector (analyzes execution traces to extract successes and failures), and SkillManager (curates the Skillbook). The Recursive Reflector writes and executes Python code in a sandboxed environment to programmatically search for patterns and isolate errors.

Key features

  • Persistent Skillbook that stores and updates strategies without fine-tuning, training data, or vector databases.
  • Three-role architecture (Agent, Reflector, SkillManager) for automated learning from execution traces.
  • Recursive Reflector that uses sandboxed Python code to analyze traces and extract actionable insights.
  • Multiple runners: ACELiteLLM, ACE, TraceAnalyser, BrowserUse, LangChain, and ClaudeCode.
  • CLI commands for setup, model search, validation, and configuration.
  • Composable pipeline engine with steps like AgentStep, EvaluateStep, ReflectStep, UpdateStep, and DeduplicateStep.
  • Proven results: doubles pass^4 on Tau2 benchmark (15 learned strategies, no reward signals), 49% token reduction in browser automation, and $1.50 learning cost for translating 14k lines of code to TypeScript with zero build errors.

Use cases

ACE is designed for AI agents that need to learn from experience. Use cases include improving consistency in multi-step agentic tasks (e.g., airline domain benchmarks), reducing token costs in browser automation, and autonomous code translation (e.g., converting Python to TypeScript). It can also learn from pre-recorded traces without re-running tasks.

Getting started

bash uv add ace-framework ace setup # Interactive setup (recommended) # Or manual configuration: export OPENAI_API_KEY='your-key' Then use in Python: python from ace import ACELiteLLM agent = ACELiteLLM(model='gpt-4o-mini') answer = agent.ask('Is there a seahorse emoji?') agent.learn_from_feedback('There is no seahorse emoji in Unicode.') answer = agent.ask('Is there a seahorse emoji?') print(agent.get_strategies())

License and pricing

The license is Apache-2.0. The project is open source. A hosted solution is available at kayba.ai.

Based on the project's public README and repository metadata.

Add this badge to your README

Featured on YunoDB Badge styles & help

Maintain this project? Show that it is featured on YunoDB.

Markdown[![Featured on YunoDB](https://yunodb.com/badge/kayba-aiagentic-context-engine.svg)](https://yunodb.com/tools/kayba-aiagentic-context-engine?utm_source=github&utm_medium=badge)
HTML<a href="https://yunodb.com/tools/kayba-aiagentic-context-engine?utm_source=github&utm_medium=badge"><img src="https://yunodb.com/badge/kayba-aiagentic-context-engine.svg" alt="Featured on YunoDB" height="20"></a>