harnyUse case
Memory for coding agents
This page describes the job memory does for a coding agent and what you build with it. It is a worked scenario, not a customer story or a measured result. A coding agent works on the same project across many sessions, so memory has to carry decisions, conventions and past mistakes forward, and tell which of them still hold.
What you are building
An agent that starts each session knowing why the project is built the way it is, what was tried and what was decided.
Why memory decides whether it works
- A new session starts blank. Decisions made last week are lost unless something carries them.
- Decisions get reversed. The agent must read the latest one, not the first.
- A record of actions helps. What the agent changed yesterday is context for today.
What Harny provides
- Ingestion of notes, decisions and session summaries with their timestamps.
- Recall that returns what is true now with dated sources.
- A record of what the agent did, kept for the next request.
- An agent setup file the agent can follow to connect itself:
harny.ai/auth.md.
Getting there
- Point the agent at
https://harny.ai/auth.mdand let it set itself up. - At the end of a session, send the decisions and open questions to
POST https://api.harny.ai/v1/memory/ingest. - At the start of the next one, call
POST https://api.harny.ai/v1/memory/recalland read the evidence first.
Frequently asked questions
Does the agent need a human to sign in first?
No. The setup guide provisions an app without an account, and gives the owner a link to claim it later.
How is this different from a notes file in the repository?
A notes file holds text. Recall returns the current value with its date and source, so a reversed decision does not get quoted as if it still stood.
Go on. Get Harny.
The first 500 monthly-active end users are free for good. Create an app key and make your first call in minutes.