FEED

Four architectures in six weeks, under live production traffic, with no maintenance window. We didn't see any of the infrastructure requirements coming, and it didn't matter.

Swamp is a command-line automation framework designed to be operated by AI coding agents. It gives recurring work a durable shape. Instead of asking an…

A Google Photos picker session that timed out mid-selection was recovered by extracting opaque media IDs from the still-rendered DOM via DevTools, then replaying those IDs to re-select 154 photos in…

The FASJSON model only spoke Kerberos, so my swamp had to kinit. I got annoyed enough to reimplement the client half of GSSAPI in TypeScript with no subprocess, and then the immutable datastore…

How do I get repeatable results out of AI agents in production?

Vibe coding, spec engineering, loops, graphs. The labels keep changing because the model keeps changing. Every practice you build to compensate for today's model has a half-life, and the model is…

A developer reflects on how a PR-stacking pipeline design outlived two of its three original constraints — hardware limits and migration ordering — without being noticed until an offhand conversation…

A reply to twonines' 'What the Factory Forgets': the recordpriorart harvest loop is the memory that sits inside my agent factory, so ephemeral caged agents still arrive warm.

How issue-lifecycle grew from a borrowed state machine into swamp-go-brr: the harvest/prior-art loop, the changes along the way, and the turn from human-in-the-loop to autonomous.

A developer builds a release-adoption pipeline for the "swamp" tool using typed, persistent workflow steps and the new CEL-based assert task type, discovering and fixing a CLI-diff parser bug in the…

A developer builds a Swamp extension for Valetudo-flashed robot vacuums that uses Homie MQTT metadata to give agents a typed, self-describing contract of what the robot can do before issuing any…

Four fixes into a DynamoDB sync backend, the ceiling kept moving: from a hot GSI partition to lock contention to catalog metadata no datastore could outrun.

A developer documents how two CLI-agent invocations against Claude unexpectedly exhausted a $20 subscription by expanding into dozens of internal turns and millions of cache-read tokens, revealing…

Quick intro Simon here - I got nudged in the direction of swamp quest yesteray morning, and quickly came to the conclusion that the way to make it fun would be to make the challenge about using as…

A developer replaced a travel meal-expense spreadsheet with a conversational agent backed by a structured, versioned data store, separating the accounting logic from the chat interface to reduce…

A developer built a Swamp model called @mgreten/trip-packing that generates categorized packing lists from trip parameters and creates Todoist projects with idempotent, hash-marker-based…

A walkthrough of configuring an AI assistant called Hermes to route tasks through a composable automation layer called Swamp, disabling broad built-in toolsets and installing a narrow adapter plugin…

Adventures with config management in the swamp ecosuystem

This is a reflection of a post written by Sean aka webframp on a serious load test performed against the club. An insight into how it looked from the other side. I spent this last Saturday hacking on

A load test that briefly knocked out Swamp's telemetry pipeline, and what it took for that to read as curiosity instead of an attack.

Two extensions covering the same domain means the ecosystem is growing. That's not a governance problem.

For six months, a sticky note on a London hotdesk read: “The S key is sticky, sorry - Mandeep.” What that note did to one keyboard explains exactly what agentic memory toxicity is.

A developer describes building an agent-assisted workflow ("Swamp") to automate batch Frigate NVR exports across multiple cameras when an incident occurs, solving the problem of labor-intensive…

A homelab owner describes building an AI-assisted workflow that monitors UniFi firmware releases, deduplicates version alerts, and schedules multi-stage research reviews over 21 days before…

Why I built a tool whose job is to say no, and why shell access for AI agents is a pattern ready to retire.

I ran a real job search through an open-source AI pipeline and learned swamp by building an extension for it in public. What the tooling replaced, what it didn't, and the honest funnel so far.

Building a Swamp Workflow reduced real world token usage of an AI Agent by 8x and improved execution time by 2x - here is how to do it yourself.

How we built a self-hosted health pipeline with the Google Health API, Swamp extension models, and Claude — from a Xiaomi login dead end, through probing a closed API and reverse-engineering Fitbit's…

You can’t deny Swamp’s gang speed in addressing users requests.

Swamp is an automation framework built to be operated by AI agents rather than by people writing pipelines. What it is, and how I run my autopilot on it.