OpenAgentlyHow it works
The local-to-cloud improvement loop
Repositories stay local at rest, retained cloud signals are redacted and derived, and every repository change waits for your approval.
Connect a project
Point the desktop app at a repository. It scans locally — languages, frameworks, structure, existing agent files — and builds a context summary. Selected, non-excluded context may be processed transiently to generate instructions, but raw source is not retained in cloud storage.
Create an agent
Describe the agent's goal and answer a few directed questions. OpenAgently generates instruction files in your harness's native format and writes them into your repo — you see exactly which files are created.
Work normally
As you use your coding agent, the local daemon captures session signals: corrections you make, errors that recur, friction that repeats. Summaries are redacted locally before they sync.
Evidence accumulates
The cloud groups signals per agent. When there's enough evidence — not before — an improvement job analyzes friction themes and gaps between the agent's goal and its behavior.
Quality gates run
Candidate instruction revisions are evaluated for safety, goal alignment, and regression risk. Only versions that pass every gate are marked ready for you.
You decide
The desktop app shows the new version with a plain-language changelog. Apply it, ignore it, or roll back later. Nothing changes in your repository until you say so.
Transparency
Every agent has an honest status
Your dashboard shows exactly where each agent sits in the lifecycle — in plain language, with timestamps.
Wondering what data actually leaves your machine?
The exact categories, retention policy, and your controls are documented in detail.
See privacy and controls