pgflow runs multi-step AI agents & jobs in your Supabase project, no extra infra.
• Declarative TypeScript with end-to-end types
• Start via TS client, RPC, pg_cron, or triggers
• Realtime progress updates
• Runs on Postgres + Edge Functions
• Production-ready, easy to use
i have growing conviction in Personal Agents. I also think they are going to be incredibly sticky
the one I'm using has started to learn my work rules/patterns & it has built up a CRM of all my contacts
it won't be so easy to shift these from one tool to another unless the industry standardizes on a way to manage handle each of these (which i doubt will happen)
i'm also surprised at how useful it has been to "call" the agent while I'm working from my computer. It's like pair-coding (for work tasks), we are both working at the same time on a kanban board to clean up a backlog of tasks
@wzulfikar i run all my mcp servers through rmcp-mux on the host - each server gets own unix socket. pi runs in a sandbox and only talks to that socket, oauth happens on the host, so tokens never enter the sandbox and no re-auth per agent 🙂
pgflow 0.16.0: one way to deploy flows. Worker startup compilation - already the default - is now the only path; the old migration system (`pgflow compile`, compileFlow, ControlPlane) is gone. Per-step queues next. Breaking: stop workers before migrating.
https://t.co/8lvvS22eta
pgflow 0.15.1: every task now reaches a terminal state — skipped steps, failed runs, stalled recovery. Double-claim protection, auto-repair of stuck rows on upgrade, and a Windows npm install fix.
https://t.co/4FE1NUW6GQ
pgflow 0.15.0: workers break out of Supabase Edge Functions. `@pgflow/edge-worker` is on npm - run workers as long-lived Node/Bun processes on Railway, Fly, Docker, or any VM. Graceful SIGTERM drain, host-managed supervision, Supabase stays the database.
https://t.co/GhT7ZdaJbI
A client trace and the matching Supabase log never shared anything, so you lined them up by guessing at timestamps.
Turn on 𝚝𝚛𝚊𝚌𝚎𝙿𝚛𝚘𝚙𝚊𝚐𝚊𝚝𝚒𝚘𝚗 and that's over. One 𝚝𝚛𝚊𝚌𝚎_𝚒𝚍 now runs from your browser straight into your Supabase logs.
Postgres Changes now supports AND filters and column selection.
You can match on more than one column in a single subscription, and pick which columns come back in the payload.
@itsmattchan that's interesting, i'm using luna xhigh/max a lot lately, as a cheapo work grunt. you guys compared on the same reasoning settings? can you tell more about what your eval tests?
@mattpocockuk The first thing i have done to grill-me is to ask questions in batches. I don't want to hit 'i agree' one by one, but I really want my answers to influence new/different questions too
@wzulfikar Good stuff! I'm on OpenAI camp and using codex load balancer from Soju06 (https://t.co/h4COSt15HZ) and i think it has lot of great ideas you can borrow, check it out
I built the open-source @supabase integration for @opencode in collaboration with the Supabase team.
It gives agents Supabase project context: orgs/projects, MCP setup guidance, bundled skills, databases, Edge Functions, logs, and more.
We built a Supabase integration for OpenCode and tested it inside Minecraft. The agent created a table, populated it with data, and deployed an Edge Function without ever leaving a survival world.
Nobody touch the dirt house. 🧱
👉 https://t.co/Tg13B6NiP9
@mattpocockuk I noticed the same thing - much easier to have all the sessions in the wiki (in my case - obsidian vault) and manage all the worktrees from there
pgflow 0.14.2: Fixes slot-aware backpressure in the edge worker - workers now respect maxConcurrent slots when polling for messages, preventing over-fetching at capacity. Also fixes shutdown handling so SQL connections are only closed when owned by the worker. Fixes #114.
@wzulfikar yup! i have overlaping use case - i'm brainstorming a lot with other people and also talk a lot and want to never forget about stuff and for important things to be surfacted (including ideas), often forgot to hit rec. was thinking a always on recording pendant or smth like that