We think the work about the work should disappear.

Every week, engineering teams lose hours to status updates, time logs, standups, and cards. Organisations lose something bigger — the record of how the work actually happened, and why. None of it is hard. It's just relentless, and by the time anyone needs the answer, the context is already gone.

But everything needed is already there, in what people just did. We build software that reads that context, understands it, and acts on it.

Small team, based in India, building in the open.

Meridian begins as a tool that logs one person's work. Where it goes is the layer an engineering organisation runs on.

Visibility, without surveillance. Managers and leaders get a live, accurate picture of what is actually moving — assembled from the worklogs each person reviewed and approved, not from raw screen capture. Capture stays on the individual's machine. What reaches the organisation is what its people chose to share, and that constraint is architectural, not a setting we can quietly change.

An organisation that understands itself. As Meridian sees work across a team, it learns the shape of the org — who owns what, how a change actually reaches production, which systems are involved and where they touch each other. Not the org chart. The way things really work.

A knowledge base, and a memory layer. Decisions, the context around them, and the reasoning behind them stop living in one person's head or a dead thread. Meridian keeps the durable record — an institutional memory the organisation can ask questions of, and the context layer that makes every AI tool in the company actually useful.

Automation that finds itself. Seeing how work genuinely gets done is what reveals the parts that shouldn't be done by hand — the recurring practices, the operational rituals, the same five steps every single time. Meridian surfaces those, and then runs them.

The end state isn't a tool anyone remembers to open. It's part of how the organisation runs.

An ambient work journal for macOS and Windows. It runs quietly in the background, reconstructs your day into real work sessions, works out which task each one belongs to, and drafts the update. Nothing posts without your approval.

Local-first: capture, OCR, and session structuring run entirely on your machine, into a database encrypted at rest. MIT licensed and fully auditable. Connects to Jira, Linear, GitHub Issues, Trello, and Azure DevOps.

Currently in early access — the core loop runs every day on real projects, and it's shipping fast.

Try it · Demo · Documentation · Source

How we think about the problem, and what we've measured while working on it.

• Your velocity went up. Your visibility went down. — why faster AI-assisted work leaves less of a record behind.
• More context made my classifier worse, not better. — a machine-maintained failure taxonomy for closing the eval loop.
• The timesheet wasn't lying. It was blind. — what a week of fully-recorded developer activity actually showed.

meridiona.com