# ace: Evolve LLM Contexts Through Agent Experience This repository profile is provided by osrepos.com, an open source repository discovery platform. Source: osrepos.com Repository profile: https://osrepos.com/repo/ace-agent-ace Generated for open source discovery and AI-assisted research. ACE helps language-agent teams build and refine reusable playbooks from task outcomes, rather than repeatedly rewriting prompts or fine-tuning models. It suits teams experimenting with offline or online adaptation for agent tasks and domain-specific reasoning. GitHub: https://github.com/ace-agent/ace OSRepos URL: https://osrepos.com/repo/ace-agent-ace ## Summary ACE helps language-agent teams build and refine reusable playbooks from task outcomes, rather than repeatedly rewriting prompts or fine-tuning models. It suits teams experimenting with offline or online adaptation for agent tasks and domain-specific reasoning. ## Topics - python - ai-agents - llm - agent-framework - context-engineering - playbook-learning ## Repository Information Last analyzed by OSRepos: Mon Oct 12 2026 17:27:23 GMT+0100 (Western European Summer Time) Detail views: 1 GitHub clicks: 2 ## Safety Notice OSRepos shares public repositories for knowledge and discovery only. Review source code, dependencies, licenses, and security implications before running or installing anything. ## Content ## Overview ACE is a Python framework for adapting language-agent context over time. It turns execution feedback into a structured playbook of strategies, insights, and mistakes to avoid, addressing the loss of useful detail that can happen when context is repeatedly rewritten. Its Generator, Reflector, and Curator roles produce experience, extract lessons, and apply incremental updates. Teams can use it for offline adaptation with training and validation data, online adaptation during evaluation, or evaluation with an existing playbook. ## Key Features - Generates task trajectories and uses feedback to identify effective strategies and recurring errors. - Separates lesson extraction from playbook curation with dedicated Reflector and Curator roles. - Applies localized playbook updates, with helpful and harmful counters, rather than replacing the full context. - Provides offline, online, and evaluation-only run modes. - Includes evaluation integrations for AppWorld, FiNER, and XBRL Formula tasks. - Supports multiple LLM API providers, including OpenAI, Together, SambaNova, and Commonstack. - Offers extension points for custom datasets, data processors, and task-specific evaluation. ## Use Cases - Agent developers can test whether accumulated experience improves performance on repeated tasks such as AppWorld interactions. - Financial NLP teams can adapt a playbook for information extraction or structured financial reasoning without labeled supervision, as described by the project. - Researchers comparing context-adaptation methods can run offline or online experiments and inspect intermediate playbooks and logs. - Teams with a custom task can implement the documented data-processing and evaluation methods to evaluate ACE on their own data. ## Our Take ACE is a promising research-oriented context-adaptation framework with an active but currently low-volume maintenance signal, and its adoption should be validated against your own tasks and API setup. **Good fit if:** - You want to experiment with evolving agent context from execution feedback instead of fine-tuning model weights. - Your work can be expressed as task data plus custom processing and evaluation, or uses one of the included evaluation tasks. - You can use Python 3.10 or later and are comfortable configuring an LLM provider API. **Look elsewhere if:** - You need a mature release cadence: the project has had no release in the past year and only a few commits recently. - You require visible recent issue resolution: no issues were closed in the recent measurement period, although a pull request was merged. - You need a packaged production service rather than a framework for running adaptation and evaluation experiments. ## Project Health | Signal | Value | |---|---| | Status | **Maintained**: a few commits in the last 90 days | | Last commit | 2026-08-24 (2 months ago) | | Commits, last 90 days | 1 | | Releases, last 12 months | 0 | | Contributors | 6 (top contributor: 42% of commits) | | Issues closed, last 90 days | 0 | | Pull requests merged, last 90 days | 1 | | Project age | 11 months | Checked on 2026-10-12 with the GitHub API. ## Project Facts - Language: Python - License: Apache-2.0 - Stars: 1.4k - Forks: 169 - Archived: no ## What You Need Detected in the repository: - Python >=3.10 (from pyproject.toml) ## Getting Started ```bash git clone https://github.com/ace-agent/ace.git cd ace uv sync cp .env.example .env ``` Configure the API key for your chosen provider in `.env`, then see the [README](https://github.com/ace-agent/ace#quick-start) for usage and run configuration. ## License in Plain Words **Apache-2.0** (permissive). - Commercial use: yes - Modify and redistribute: yes - You must keep: the license, the NOTICE file and a note of your changes - Share your changes: no - Includes an explicit patent grant from the contributors A summary, not legal advice: the LICENSE file is what applies. ## Considerations - The measured status is Maintained, but recent activity is limited to a few commits, with no release in the past year. Six contributors have participated, and contribution is not concentrated in one person. - The repository reports no CI configuration or test suite in the measured facts. Teams adopting it should assess how they will validate changes and their own experiment results. - Running the framework requires access to a supported LLM API provider and appropriate credentials. The README does not specify a GPU or database requirement. - ACE is Apache-2.0 licensed. Review the license terms when redistributing or incorporating it into a product. - The README lists extending ACE for tool calling as coming soon, so that capability is not documented as an available tutorial or feature.