Architecture decisions, customer patterns, and the occasional rant from the team building Evoxiv. Roughly two posts a month, no AI-written filler.
A flat markdown file, capped at 4,000 characters, and a single continuation run after every Story. Three rules that turn an Agent from forgetful to compounding — without a vector database in sight.
A flat markdown file, capped at 4,000 characters, and a single continuation run after every Story. Three rules that turn an Agent from forgetful to compounding — without a vector database in sight.
The case for inverting the assignee. Why a "human assignee" field would have dissolved Evoxiv into Linear with an AI button — and how we kept the constraint load-bearing.
No per-seat tier-jumping. No usage rails. Three plans with one shape — and the math behind why that works when you don't resell tokens or run the daemon.
We launched Evoxiv with cronjobs in v0.1 because recurring Stories are how Agents earn their keep. Here's how we model schedules, templates, and per-tick lineage.
A walkthrough for running `packages/daemon` as a Cloud Run job, with autoscaling on task-queue depth and graceful drain on revision rollover.
Six features we shipped, lived with, and then deleted: comments, labels, priorities, a notifications inbox, an "@mention" parser, and a per-user dashboard. A love letter to the constraints that survived.
We talked to 23 teams running Agents in production. Here's what they said about Memory, scheduling, and why they kept rebuilding the same internal tool until they found ours.
Roughly two posts a month — architecture, customer interviews, the occasional retro. No drip campaigns, no AI summary digests, unsubscribe in one click.
