OpenAI builds and deploys advanced AI models like GPT-4o for autonomous agents and workflows.
Autoexp
0Popularity
About Autoexp
Autoexp provides reproducible experiment boundaries for coding agents, recording exact source code, inputs, logs, outputs, metrics, reports, and lineage. It offers a browser-based review surface for human feedback and preserves evidence for later inspection. The tool integrates with agents like Codex, Claude Code, OpenCode, and Pi, creating a structured loop where agents run experiments, Autoexp seals the results, and humans review and steer the next attempt. Users invoke /autoexp to start an experiment, which the agent executes within the existing repository. Autoexp then preserves all artifacts and allows human review via /autoexp-review, where feedback is submitted and returned to the waiting agent as the next instruction. A companion command, autoexp view, provides a read-only dashboard to browse and download experiment history without feedback capability.
Key features
- Reproducible experiment boundaries
- Preservation of source, inputs, logs, outputs, and lineage
- Browser-based review surface for human feedback
- Integration with Codex, Claude Code, OpenCode, and Pi
- Rerun and compare exact experiment attempts
- Local dashboard for browsing experiment history
- Structured feedback loop between human and agent
- Apache 2.0 licensed
Use cases
- Running and comparing AI-led experiments within a codebase
- Reviewing and steering agent-generated work before finalization
- Preserving and analyzing experiment history for research projects
Pros
- Preserves exact source, inputs, logs, outputs, and lineage per experiment
- Provides a browser-based review surface for human feedback
- Integrates with multiple coding agents (Codex, Claude Code, OpenCode, Pi)
- Supports reproducible reruns and comparisons of exact attempts
- Local-first with no cloud dependency
Cons
- Requires specific coding agents for full functionality
- No cloud-based collaboration features
- Limited to experimentation tasks; not for ordinary edits
- Agent must pause during review sessions
Frequently asked questions about Autoexp
What does Autoexp do?
Autoexp provides local-first experimentation infrastructure for coding agents, creating reproducible experiment boundaries and recording exact source code, inputs, logs, outputs, metrics, reports, and lineage. It offers a browser-based review surface for human feedback and preserves evidence for later inspection.
Who is Autoexp for?
Autoexp is designed for developers and teams using AI coding agents such as Codex, Claude Code, OpenCode, and Pi. It suits those who need to run, track, and review agent-led experiments while maintaining reproducibility and structured feedback loops.
How do I get started with Autoexp?
Install Autoexp using the provided script for your operating system, then restart your coding agent. Use the /autoexp command to start an experiment or /autoexp-review to inspect and provide feedback on existing results.
Does Autoexp integrate with specific coding agents?
Yes, Autoexp integrates with agents like Codex, Claude Code, OpenCode, and Pi. The installation process creates the necessary plugin or extension for each agent, enabling seamless interaction.
Can I use Autoexp for non-experimental tasks like ordinary edits?
No, Autoexp is intended for reproducible runs, experiment history, or tasks requiring structured feedback. It should not be used for ordinary edits that do not need preserved runs or experiment tracking.
How does the feedback loop work in Autoexp?
After an agent completes an experiment, Autoexp pauses and opens a browser-based review session. Users can inspect evidence such as source code, logs, and metrics, then submit feedback. This feedback is returned to the waiting agent as the next instruction.
