Watches your codebase.
Icarus reads merged PRs, closed issues, and load-bearing commits as they land, and distills them into clean Markdown memory. The team's decisions become the agent's working knowledge, automatically.
Marks what changed.
Stale facts are superseded, contradictions are surfaced, and every memory links back to its source. You see what the team believed last quarter, what they believe now, and which arguments were settled in between.
Briefs your agents.
Before a coding agent starts work, Icarus gives it a short sourced briefing of what is current, what is stale, what is risky, and what has failed before. Claude, Cursor, and your own agents all read from the same memory.
The team's memory should not live in the team's heads.
A memory entry.
Every fact in Icarus is a plain Markdown file. Status, sources, and a last-verified date in the frontmatter. The body in your team's voice. Nothing about it is opaque.
We chose Postgres in March 2025 after evaluating Cloud Spanner and CockroachDB. The deciding factors were operational familiarity and the team's existing migration tooling. Considered and rejected: - Spanner: cost prohibitive at our scale. - CockroachDB: added consistency latency the product surface didn't need. Revisit if write volume passes ~50k tps or we need true multi-region writes.
Common questions.
How does Icarus get its information?
It reads the activity your team already produces on GitHub. Merged PRs, closed issues, load-bearing commits. Every memory entry cites the artifacts it was built from, so your team can verify any claim it makes.
Where does the memory live?
In your repo, as plain Markdown files in a directory you choose. Version-controlled with your code. If you walk away from us tomorrow, you keep everything. There is no proprietary format to escape from later.
Which agents can read it?
Anything that speaks the Model Context Protocol. Claude Desktop, Cursor, Continue, internal agents you build yourself. The same memory feeds all of them, so your tools agree on what is true.
Does Icarus write to my code?
No. Icarus opens pull requests against your memory directory only. Your team reviews them like any other change. Your application code is never touched by Icarus.
Is my source code sent anywhere?
Source stays in your repo. The summarization step uses the language model provider you configure with your own key. We never see your code. We never store it.
Book a pilot
Icarus is in private beta, built first for coding agents on long-lived codebases. Drop your email. We'll be in touch.