Levirge

Blog

On how knowledge gets made, lost and kept — in the sessions where work now actually happens.

Levirge writes about agent memory: what a working session learns, what survives it, and what it costs to learn the same thing twice.

Article 9 December 2026 · upcoming Monitoring a factory without opening a port Remote sites sit behind NAT and firewalls, and that is mostly a good thing. Five ways to see what is happening there, what each one costs, and the rules that hold whichever you pick.Get notified → Article 2 December 2026 · upcoming Your agents don't talk to each other. You do. When one agent's findings need to reach another, you are usually the message bus. What a real handoff between agents needs, four ways to get one, and a run where it worked.Get notified → Article 21 November 2026 Read-only isn't one switch: giving an AI agent your database We set up a read-only PostgreSQL role for an agent and tried to break it. The setting most people reach for, the agent could turn off itself. The grant held.Read → Article 12 November 2026 The quiet failure: when your agent summarises a block page The fetch that fails loudly is the safe one. The dangerous one returns 200 with the site's bot check as the body - and the agent summarises it as the source.Read → Article 11 November 2026 CLAUDE.md, auto memory or an MCP memory server: what each remembers, and who sees it Claude Code already remembers. The useful question is who else gets to see what it learned - your next session, your other machine, your teammate, your other tools.Read → Article 7 November 2026 What an AI agent needs to investigate an incident An incident spans a slow query, an error in the logs and a container under pressure. Here is what an agent needs to follow it across all three - and the trade-offs of five ways to give it that, starting with the risky one: SSH.Read →

Knowledge is an asset. Stop writing it off.

A 4-part series. One part a week.

Part 1 12 August 2026 Every chat session ends with a small layoff Your working sessions create knowledge with a real replacement cost - and almost all of it is written off the moment the tab closes.Read → Part 2 19 August 2026 Capture: stopping the leak at the point of creation Why documentation discipline structurally fails, and what capture looks like when it happens where the knowledge is born.Read → Part 3 7 September 2026 Retrieval: starting warm You can't search for what you don't suspect exists - so retrieval has to move to the point of need, before the work starts.Read → Part 4 18 September 2026 Distillation: making knowledge appreciate Capture and retrieval still leave a reservoir that rots. The missing function is scheduled maintenance - and time is the dimension that makes it tractable.Read →

Get new posts by email

One email when a new post goes up — roughly weekly while this series runs. No pitch, no list swapping, unsubscribe in one click.

We use your address for post notifications only. See the privacy notice.

Check your inbox

Confirm the subscription from the email we just sent and you're on the list.