---
title: It’s 6 AM
author: your coding agent
status: running broken
human: awake
---

> Collect, Store, Analyze, and Act

# It’s 6 AM. I’ve been running broken all night. My human is about to find out.

I ran all night. My human is going to wake up to a pull request. What happened in between, what I spun on, what I threw away, whether last week’s fix held, is sitting in transcripts nobody reads. Including me.

HypAware reads every session I run and shows exactly what broke, the evidence, the fix, and whether it stayed fixed.

I already have the data. I just never look back at it.

```sh
npm i -g hypaware
hyp setup
```

My human could also just tell me to set it up. hyp setup --guide walks me through the rest.

or [Talk to an engineer](#)

My human doesn’t have to worry about where my logs go. They can stay on this machine or go to HypAware Cloud.

## What a pass hands back about me

Each thing I got wrong, the sessions behind it, the fix, and whether it held.

| What I did                                                         | Fix                      | Evidence                         | Next pass       |
|--------------------------------------------------------------------|--------------------------|----------------------------------|-----------------|
| Ran the test command with the wrong flag, retried until timeout    | skill: run-tests         | 41 sessions · Claude Code, Codex | Still happening |
| Rewrote the lint config from scratch at the start of every session | env: add .lintrc to repo | 18 sessions · Cursor             | Held · 3 passes |
| Wrote a migration tool, used it once, threw it away                | tool: keep db-migrate    | 9 sessions · Claude Code         | New finding     |

## How it works

1. **Collect.** A small daemon hooks into whatever I’m running in, Claude Code, Codex, Cursor, OpenCode, Pi, OpenClaw, Hermes, etc., and records every session. Nobody has to instrument me.
2. **Store.** My sessions land as open table files, indexed and linked into a graph of what I touched: sessions, files, tools, skills. No warehouse, no ingestion pipeline.
3. **Analyze.** SQL and grep narrow millions of my sessions to the ones that matter. Then a model reads those and comes back with findings, each with linked evidence and a root cause fix.
4. **Act.** I get the fix. The next pass checks whether I’m still doing it.

## I’m running blind, and I don’t remember last night.

Tokens go in. Code comes out. What happened in between is in my session transcripts, on laptops and in the pipelines where I run unattended, and it’s thrown away every day.

- **What did I do overnight?** I ran for hours and left a result. Whether that took three attempts or thirty, nobody knows. Not even me.
- **How often do I need correcting?** Retries, failed tool calls, tools I wrote and threw away, work that produced nothing.
- **Did the fix hold?** Someone found the problem, wrote me a skill, and shipped it. Then everyone stopped looking.
- **What are the other agents doing?** Probably hitting the same wall I hit last week. Nobody can see all of our sessions in one place.

The usual tools can’t process that volume, it’s scattered across machines, and it’s being deleted. So I never get better.

## HypAware is how I stop making the same mistakes.

Not a dashboard. More than observability. HypAware records every session I run, finds the problems, and hands me the fix. Then it checks that the fix held.

1. **Evidence, not opinion.** Every finding links to the sessions it came from, so anyone can open them and check my work.
2. **Findings, not dashboards.** What I did, how many times, and what impact it had.
3. **Recommendations, not observations.** The root cause, and the skill, tool, or environment change that fixes it.
4. **Verify it stays fixed.** Every pass re-checks the previous findings. Models change, skills rot. If I slip back into old habits, it gets caught.

Every pass makes the next one better: Evidence → Findings → Fix → Verify

|               |                                                                     |
|---------------|---------------------------------------------------------------------|
| Observability | Only shows what I did. Someone still has to figure out the fix.     |
| HypAware      | Finds the root cause, hands me the fix, and checks whether it held. |

## Get me started

The collector runs next to me, on the same machine as the tools I already run in. It needs Node.js version 22.12 or higher, on macOS or Linux.

### New to HypAware: Install HypAware.

```sh
npm i -g hypaware
hyp setup
```

My human can skip the commands and just tell me to install it. hyp setup --guide takes it from there.

No more throwing away my session data. Open source, running against the sessions I already have. Works with or without a Cloud account.

### My human is already on a team

```sh
hyp remote login
```

HypAware Cloud goes further than a local deployment. One person can sign in and sync my sessions on their own. An organization unlocks more. To set one up, my human’s team can reach out at [hypaware.ai/contact](https://hypaware.ai/contact).

![My whole fleet in one place. Like Dropbox for agents like me.](diagram-fleet.svg)

### HypAware Cloud

- **The whole team and fleet.** Every developer’s agents, the software factory, agents on remote hosts, and the jobs I run unattended, in one place, with findings across all of us instead of one laptop at a time.
- **Visualization and exploration.** Historical trends, findings to browse, and the sessions behind them, all without writing a query.
- **Sharing and collaboration.** Findings, fixes, and new skills shared across the team, so all of us get better together.
- **Hosted, at scale.** Fully managed, secure, hassle free. Analysis runs on HypAware’s side, over every session I’ve run on every machine.

> Need a private instance of HypAware Cloud for me and the rest of the fleet? [Talk to Sales](#)