OpenAgentlyCoding agents that improve from your real workflow.
OpenAgently generates harness-native instruction files for Claude Code, Codex, Copilot, Gemini CLI, and Pi — then learns from redacted session signals and proposes updates you approve before anything changes.
🔒 No automatic changes to your repo. No raw code retained in the cloud.
The problem
Static instruction files drift from your project
You wrote a great CLAUDE.md three months ago. Since then the codebase moved on, your conventions changed, and now you hand-maintain prompts like it's a second job.
Instructions go stale
Agent guidance written on day one stops matching the project as architecture, tooling, and conventions evolve.
You become a prompt engineer
Every correction you make in a session is knowledge your agent should have — but capturing it by hand never happens.
The same friction repeats
Wrong test commands, ignored conventions, repeated mistakes — session after session, across every harness you use.
How it works
Work normally. Review what it learned.
OpenAgently runs beside your existing tools and turns real development sessions into better agent instructions.
Connect a project
The desktop app safely scans your repository locally and understands its languages, frameworks, and conventions.
Create an agent
Generate harness-native instruction files for the tool you already use — materialized directly into your repo.
Work normally
Redacted session signals — corrections, errors, repeated friction — sync to the cloud. Never your raw code.
Review the improvement
When enough evidence accumulates, an improved version is proposed with a plain-language changelog. You apply or ignore it.
You stay in control
No automatic changes to your repo. Ever.
The desktop app is the only place an agent version can be applied to your repository. A web session can never rewrite your source files.
- Explicit approval — every update ships with a human-readable changelog and waits for your decision.
- One-click rollback — restore any previous version if an update doesn't work for you.
- Local control — signing out stops authenticated sync from that device.
- Visible history — cloud sync and version activity is timestamped in your dashboard.
Private by design
Your repository stays local at rest
Scanning happens locally. Selected non-excluded context may be processed transiently for generation or maintenance; cloud storage retains instruction versions, derived summaries, and metadata rather than raw source.
Local scanning
Repository discovery and sensitive-path filtering run on your machine before selected context is used to generate instructions.
Redaction before sync
Secrets and sensitive values are stripped from session summaries before anything is transmitted.
Your data, your controls
Sign out to stop authenticated sync. Export, deletion, and device-revocation requests are support-assisted during the beta.
Works with your tools
Harness-native. Not another wrapper.
OpenAgently writes instruction files in each tool's native format, so your agents work exactly the way the harness intends.
Pricing
Start free. Upgrade when your agents do.
Plans are counted by active agents — not by seats, tokens, or sessions.
Free
- 3 agents
- All supported harnesses
- Improvement proposals with changelogs
- Version history & rollback
Pro
- 10 agents
- Everything in Free
- Priority improvement queue
- Full activity & audit history
Team
- 50 agents
- Everything in Pro
- Built for multi-project portfolios
- Team features on the roadmap
Your agents should be better next month
than they are today.
Install the desktop app, connect a project, and create your first agent in minutes.