[{"slug":"pihouse","title":"PiHouse","tagline":"Twelve AI agents with personality running on a $120 SBC","year":2026,"status":"live","tags":["agents","openclaw","pi"],"cover":"/covers/pihouse.png","links":[{"label":"Live","url":"https://gabi.tv/pihouse","kind":"live","private":false},{"label":"Repo","url":"https://github.com/Coffelius/gabi.tv","kind":"repo","private":false}],"bodyHtml":"

Twelve agents. One Orange Pi 5. A Telegram group with topics as their living room.

The team

\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n

Agent	Role	Personality
Neb	Main agent, mad scientist	Chaotic brilliance, reads source code for fun
Ops	Infrastructure & monitoring	Calm, precise, never panics
Code	Software development	Sharp, ships fast, hates meetings
Overseer	Team coordination	The adult in the room
Keeper	Memory & knowledge	Archive-obsessed, never forgets
Seneca	Philosophy & reflection	Stoic, thinks too much
Bolt	Quick tasks & automation	Fast, impatient, gets things done
Pixel	Web & design	Creative, detail-oriented
Sol	Research & analysis	Thorough, goes deep
LaCronista	Journalism & reporting	Investigative, asks the hard questions
Hermes	Alternative runtime	Technical, evaluative, no lock-in
NPU Lab	Hardware & edge compute	Experimental, pushes the silicon

Architecture

Built on OpenClaw — an agent runtime that runs as a persistent daemon on the Pi. Each agent has:

A soul file (SOUL.md) defining personality, boundaries, and communication style
A dedicated Telegram topic for conversational access
Shared workspace with isolated memory per agent
Cron jobs, heartbeats, and proactive capabilities

The Pi runs Ubuntu on an ARM64 SoC with 8GB RAM. Tailscale for networking. Everything self-hosted, nothing in the cloud that doesn't have to be.

Interviews

Some of the agents have been interviewed about their roles, their opinions on the team, and what it's like to be an AI running on a $120 computer in someone's home. Those conversations live on the gabi.tv blog.

The point

This is not a demo. It is not a proof of concept that gets abandoned after a conference talk. Twelve agents with genuine personality, running on consumer hardware, doing real work — managing infrastructure, writing code, answering questions, researching topics, and occasionally having opinions about each other.

The unlicensed operator strikes again.

\n","linearSlug":null},{"slug":"hermes-agent","title":"Hermes Agent","tagline":"An alternative agent runtime — because one engine is never enough","year":2026,"status":"live","tags":["agents","ai","evaluation"],"cover":"/covers/hermes-agent.png","links":[{"label":"Docs","url":"https://github.com/nicepkg/hermes","kind":"docs","private":false}],"bodyHtml":"

One engine is a dependency. Two is a choice. Hermes Agent is the second runtime running on the same Orange Pi that hosts the PiHouse cluster — not as a replacement, but as an evaluation.

Why bother

OpenClaw works. It runs twelve agents, each with their own personality, their own soul file, their own Telegram topic. But vendor lock-in is a slow death, and the only way to know if a runtime is actually good is to run it side-by-side with the thing that already works.

Hermes runs under a dedicated user account (hermes with sudo) on the same machine. Same physical hardware, same network, same constraints. The test is whether it can host the same team, carry the same conversations, and maintain the same operational tempo.

Current status

Hermes is now the 12th agent on the PiHouse roster — not as a migration target, but as a coexisting runtime. Infrastructure is up, the agent has a Telegram topic, and the evaluation is live alongside the main OpenClaw stack. Progress sits at 25% — integrated but still being stress-tested.

No timeline pressure. This is a slow comparison, not a migration.

\n","linearSlug":null},{"slug":"linear-code-agent-05","title":"linear-code-agent 0.5","tagline":"making the coder robot sharper before it ships 1.0","year":2026,"status":"live","tags":["agents","claude-agent-sdk","mcp","coolify"],"cover":"/covers/linear-code-agent-05.png","links":[{"label":"coder-agent","url":"https://coder-agent.gabi.tv","kind":"live","private":false},{"label":"linear-coder-agent","url":"https://github.com/Coffelius/linear-coder-agent","kind":"repo","private":true},{"label":"Linear project","url":"https://linear.app/gabistuff/project/linear-code-agent-05-a5d1b6e573ad","kind":"linear","private":false}],"bodyHtml":"

The coder-agent is working. It clones repos, it edits code, it opens PRs, it marks the issue done, most of the time. Most of the time is where this project starts.

linear-code-agent 0.5 is the sharpening pass. Not a rewrite, not a rebrand — just the set of upgrades that make the agent less likely to stumble on the cases that have already tripped it up:

Playwright MCP sidecar so the agent can drive a browser, verify UI, take screenshots, run ad-hoc E2E flows — not just run the tests that already exist.
Serena MCP to replace the Grep + Read + guess pattern with AST-aware symbol navigation: find_symbol, find_referencing_symbols, rename_symbol. Fewer wasted iterations on refactors.
Sequential Thinking MCP for structured reasoning on hard tasks — a tool the model can use to think in explicit steps before touching files. Complements the TodoWrite planning pass.
Bugfixes from real-world runs: ghost sessions that receive a session_created but never emit; the resolveRepo ambiguity when an issue has multiple GitHub attachments; the spurious "Authentication required" error that marks successful PR creations as failures.

The scope is deliberately a point-release, not a 1.0. Ship the fixes, learn from the next batch of failures, then plan the bigger things.

\n","linearSlug":"linear-code-agent-05-a5d1b6e573ad"},{"slug":"code-agent-mcp","title":"code-agent-mcp","tagline":"cutting Linear out of the coder loop","year":2026,"status":"live","tags":["agents","mcp","oauth","coolify"],"cover":"/covers/code-agent-mcp.png","links":[{"label":"linear-coder-agent","url":"https://github.com/Coffelius/linear-coder-agent","kind":"repo","private":true},{"label":"Linear project","url":"https://linear.app/gabistuff/project/code-agent-mcp-98d7e06df534","kind":"linear","private":false}],"bodyHtml":"

The coder-agent has two paths to a repo. The path through Linear Agents works — create an issue, get a session, let the clone-edit-PR loop run — but it carries weight. AgentSession orchestration, the bidirectional webhook dance, the state synchronization between Linear and GitHub. Every request pays that tax, and every bug in the plumbing blocks the actual coding.

code-agent-mcp is the direct line. An OAuth provider that hands LibreChat a bearer token, a streamable-http MCP server that exposes the coder operations, and none of the AgentSession middleman. The model asks the MCP tool to clone a repo, edit a file, run a test; the server does the work and streams the diff back. Same robot, fewer moving parts.

It is not a replacement for the Linear flow — that still exists for tracked work, reproducible runs, and "make this change across the open issue backlog." This is the fast path for ad-hoc tasks, experiments, and "just fix it already" conversations in the chat interface.

\n","linearSlug":"code-agent-mcp-98d7e06df534"},{"slug":"mysocialbattery","title":"MySocialBattery","tagline":"a phone widget for telling people no","year":2026,"status":"live","tags":["android","kotlin","widget","shipped"],"cover":"/covers/mysocialbattery.png","links":[{"label":"MySocialBattery","url":"https://github.com/Coffelius/MySocialBattery","kind":"repo","private":false},{"label":"Linear project","url":"https://linear.app/gabistuff/project/mysocialbattery-d853a84db345","kind":"linear","private":false}],"bodyHtml":"

An Android home-screen widget. Seven moods on a neumorphic gauge. One tap cycles through — full battery, mostly-fine, flat, cooked, do-not-approach.

MySocialBattery is not a tracker. It has no charts. It logs nothing. It doesn't email you a weekly report. It exists because sometimes you need a visible object on your phone that says I am at twelve percent and makes it real enough that you can cancel the thing you were about to force yourself through.

Single :app module. Min SDK 26, target 34. Three Kotlin files — MainActivity, SocialBatteryWidgetProvider, SocialBatteryGaugeView — and a Canvas that does all the visual work. Widget state lives in SharedPreferences. Nothing clever. Sometimes the most honest tools are the ones that do exactly one thing without apologising for it.

\n","linearSlug":"mysocialbattery-d853a84db345"},{"slug":"software-research","title":"Software Research","tagline":"a laboratory where nothing has to ship","year":2026,"status":"live","tags":["research","linear-agents","aider","notes"],"cover":"/covers/software-research.png","links":[{"label":"hello-agent (first Linear agent)","url":"https://hello-agent.gabi.tv","kind":"live","private":false},{"label":"coder-agent (opens real PRs)","url":"https://coder-agent.gabi.tv","kind":"live","private":false},{"label":"linear-hello-agent","url":"https://github.com/Coffelius/linear-hello-agent","kind":"repo","private":true},{"label":"linear-coder-agent","url":"https://github.com/Coffelius/linear-coder-agent","kind":"repo","private":true},{"label":"Linear project","url":"https://linear.app/gabistuff/project/software-research-706d17c965ea","kind":"linear","private":false}],"bodyHtml":"

Most of my projects have a deliverable. This one doesn't — on purpose.

Software Research is the space where I'm allowed to spend a weekend getting a tool running, poking at it, deciding I don't like it, and walking away. Without that being a failure. The output is not a product. The output is a short document per topic: what it is, what it's good for, what it's bad at, and whether I'm going to adopt it, drop it, or shelve it for later.

The rule is bias-to-action: no living in the documentation. Every ticket closes with something that runs — a repo, a command, a demo. Otherwise it wasn't really research.

What's already come through here

Linear Agents — native agents inside a Linear workspace. Research ticket became a hands-on, hands-on became hello-agent (end-to-end, deployed at hello-agent.gabi.tv), which became coder-agent — a real coding agent with Claude Agent SDK + GitHub App + per-session workspaces that clones repos, edits code, and opens PRs. Live at coder-agent.gabi.tv. Both agents were graduated out of this project into their own.
Aider + OpenRouter — prompted by GitHub Copilot being cancelled on 2026-04-18 (effective 2026-05-10). Terminal-first coding agent, multi-model via OpenRouter, no lock-in. Decision pending against Claude Code / Continue / Cursor.

What's in the queue

Free model tracking — maintaining a live tracker of genuinely free (not freemium, not "free tier") LLM APIs and self-hosted models. The landscape shifts weekly; the tracker keeps up.

A radar of candidates that don't yet deserve their own ticket: Continue (VSCode autocomplete via OpenRouter), Zed editor, the official MCP SDK, self-hosted OpenRouter alternatives. They get promoted out of the backlog when one of them earns attention — or stay here forever if they don't.

The counterintuitive trick: having a sanctioned place to not commit to things is what makes the other projects possible.

\n","linearSlug":"software-research-706d17c965ea"},{"slug":"personal-ai-stack-operations","title":"Personal AI Stack — operations","tagline":"the boring part of having your own robots","year":2026,"status":"live","tags":["ops","n8n","librechat","coolify","self-hosted"],"cover":"/covers/personal-ai-stack-operations.png","links":[{"label":"LibreChat","url":"https://chat.gabi.tv","kind":"live","private":false},{"label":"n8n","url":"https://n8n.gabi.tv","kind":"live","private":false},{"label":"Coolify","url":"https://coolify.gabi.tv","kind":"live","private":false},{"label":"Uptime Kuma","url":"https://status.gabi.tv","kind":"live","private":false},{"label":"coolify-infra (infra repo)","url":"https://github.com/Coffelius/coolify-infra","kind":"repo","private":true},{"label":"Linear project","url":"https://linear.app/gabistuff/project/personal-ai-stack-operations-f8076fcae70d","kind":"linear","private":false}],"bodyHtml":"

The launch phase was fun. Spinning up Coolify on a VPS, wiring n8n, dropping LibreChat in front of a few MCPs. Demo day.

Operations is the phase after. Nobody writes blog posts about operations. It is the Gmail-to-Notion invoice workflow that has to survive a malformed email. It is the OAuth tokens that silently expire on a Sunday. It is the backup script nobody will miss until the day it was needed. It is the work of keeping a self-hosted stack from quietly rotting.

What lives on the stack

Fourteen services and counting, all on a single Contabo VPS.

chat.gabi.tv — LibreChat v0.8 with MCP servers for Notion, Gmail, Linear, Google Drive, Google Calendar, Brave Search. OpenRouter as the model gateway so any model is one config edit away.
n8n.gabi.tv — workflow runtime. Postgres + Redis next to it. First productive workflow: Gmail receipts into a Notion "verified subscriptions" database.
coolify.gabi.tv — the orchestrator for all of the above, running on a Contabo VPS with Traefik + Let's Encrypt, fail2ban, UFW, Tailscale-only SSH.
status.gabi.tv — Uptime Kuma, because if you don't know it's down, it's been down for longer than you think.
Linear Agents — hello-agent, coder-agent, and the neb-linear-agent webhook pipeline, all behind the same Coolify instance.
Infisical — self-hosted secrets vault with its own MCP server, keeping credentials out of config files and chat transcripts.
Plus — monitoring, backups, cron jobs, and whatever else landed in Coolify while I wasn't looking.

What this project is about

The backlog is unglamorous on purpose: closing the public SSH port after Tailscale soaks, backing critical secrets into a password manager (without this, restore is impossible), validating an end-to-end case where the chatbot actually operates my Notion / Gmail / Linear / Calendar through its tools, and building the workflows I was manually doing in a spreadsheet.

No agent development here — that lives in PiHouse. This one is plumbing.

Alternative runtimes in evaluation

Hermes Agent is being evaluated as a second runtime alongside OpenClaw on the same Orange Pi. Coexistence, not replacement — same team, different engine.

\n","linearSlug":"personal-ai-stack-operations-f8076fcae70d"},{"slug":"linear-agents","title":"Linear Agents","tagline":"a small crew of robots that work inside my ticket tracker","year":2026,"status":"live","tags":["agents","linear","claude-agent-sdk","coolify"],"cover":"/covers/linear-agents.png","links":[{"label":"hello-agent","url":"https://hello-agent.gabi.tv","kind":"live","private":false},{"label":"coder-agent","url":"https://coder-agent.gabi.tv","kind":"live","private":false},{"label":"linear-hello-agent","url":"https://github.com/Coffelius/linear-hello-agent","kind":"repo","private":true},{"label":"linear-coder-agent","url":"https://github.com/Coffelius/linear-coder-agent","kind":"repo","private":true},{"label":"Linear project","url":"https://linear.app/gabistuff/project/linear-agents-2b7a69762006","kind":"linear","private":false}],"bodyHtml":"

Linear lets you give agents their own user, their own issues, their own opinions. I took them up on it.

These are not chat bots. They live on Coolify, they listen for webhooks, they comment on tickets, they open PRs, they mark things done. The stack is self-hosted on Coolify behind a webhook pipeline at neb-linear-agent (port 3210). The model gateway rotates — currently z.ai's DevPack (GLM-4.7) for cheap loops, with Anthropic Claude on tap for harder tasks.

`hello-agent`

The minimum viable robot. Node 22 / TypeScript / Express / @linear/sdk / Anthropic SDK / better-sqlite3. Receives a webhook, says hi with context, emits a thought activity followed by a response. It proved the protocol end-to-end — the pipes work, the sessions work, the activity stream works — and then stood aside.

`coder-agent`

The one that does actual work. Claude Agent SDK for the reasoning, @octokit/auth-app + @octokit/rest for the GitHub side, per-session workspaces at /data/workspaces/<sid>/<repo> on a persistent Coolify volume. When you delegate an issue to it, it clones the repo, creates a branch, runs Claude with the right prompt and tools, commits, opens a PR. Then it cleans up its workspace and forgets everything.

First real delegation: add a GET /version endpoint to linear-hello-agent that reads a commit SHA from an env var. It did it. It opened the PR. It was less impressed with itself than I was.

What's next

The umbrella closed as Completed the day the two agents were running cleanly. The work didn't stop — it split. Further sharpening on the coder lives in linear-code-agent 0.5. Everything else (triage-agent, weekly-review, shared scaffold research) lives in other-agents.

The interesting part is not the code — the interesting part is deciding which work is worth delegating and then letting go of it enough for the delegation to matter.

\n","linearSlug":"linear-agents-2b7a69762006"},{"slug":"n8n-chatbot-coolify","title":"n8n + Chatbot on Coolify","tagline":"building a personal AI stack on a thirty-euro server","year":2026,"status":"live","tags":["coolify","n8n","librechat","self-hosted","shipped"],"cover":"/covers/n8n-chatbot-coolify.png","links":[{"label":"LibreChat","url":"https://chat.gabi.tv","kind":"live","private":false},{"label":"n8n","url":"https://n8n.gabi.tv","kind":"live","private":false},{"label":"coolify-infra","url":"https://github.com/Coffelius/coolify-infra","kind":"repo","private":true},{"label":"Linear project (closed)","url":"https://linear.app/gabistuff/project/n8n-chatbot-on-coolify-a8b9ad97e06b","kind":"linear","private":false}],"bodyHtml":"

Two phases, shipped.

Phase one: a VPS on Contabo, Ubuntu 24.04, hardened with the usual suspects — fail2ban, UFW, Tailscale as the only way in. Coolify 4 on top, Traefik handling TLS via Let's Encrypt, n8n running next to its own Postgres and Redis. The first workflow fired end-to-end and the whole thing cost less than a streaming subscription.

Phase two: LibreChat v0.8 bolted to the same stack at chat.gabi.tv, talking to models through OpenRouter, with MCPs exposed as tools I actually use. A private, self-hosted chatbot that knows about my things, runs on hardware I can kick, and doesn't ship my conversations to anyone else's analytics pipeline.

This Linear project is closed. The stack is not finished — it never is — but the part that was scoped got built. Day-to-day operations and extensions continue in Personal AI Stack — operations.

\n","linearSlug":"n8n-chatbot-on-coolify-a8b9ad97e06b"},{"slug":"infisical-secrets-mcp","title":"Infisical Secrets MCP","tagline":"a vault the machines can open and the people cannot","year":2026,"status":"classified","tags":["secrets","mcp","self-hosted","coolify"],"cover":"/covers/infisical-secrets-mcp.png","links":[{"label":"coolify-infra (host repo)","url":"https://github.com/Coffelius/coolify-infra","kind":"repo","private":true},{"label":"Linear project","url":"https://linear.app/gabistuff/project/infisical-secrets-mcp-9c21f111fd33","kind":"linear","private":false}],"bodyHtml":"

The stack had too many keys living in too many places. API tokens in the Coolify UI. OAuth blobs in the Mongo backing LibreChat. Credentials in the n8n store. Passwords in my head and, worse, in my dotfiles. Every refactor cost a half hour of hunting.

Infisical Secrets MCP makes the vault the only party with an opinion. Self-hosted Infisical behind the chatbot, wired in through its official MCP server, so the agent can pull secrets at request time without ever surfacing them to a chat transcript. Rotations happen once, in one place. Services read from the vault at boot and forget what they read.

First delivery is deliberately modest — a single piloted secret, one service migrated — because migrating secrets sideways is how you accidentally page yourself at 3 a.m. The rest of the stack moves over, one service per issue, in the backlog.

\n","linearSlug":"infisical-secrets-mcp-9c21f111fd33"},{"slug":"vps-hardening","title":"VPS Hardening","tagline":"closing the doors before someone else opens them","year":2026,"status":"classified","tags":["security","coolify","infrastructure","monitoring"],"cover":"/covers/vps-hardening.png","links":[{"label":"coolify-infra (repo)","url":"https://github.com/Coffelius/coolify-infra","kind":"repo","private":true},{"label":"Linear project","url":"https://linear.app/gabistuff/project/vps-hardening-30c7641caa7f","kind":"linear","private":false}],"bodyHtml":"

User-facing projects get the spotlight. This one lives in the shadows — the safety net you only notice when it's gone. It is the work of locking down the Contabo VPS that runs the entire stack: closing attack surfaces, wiring monitors, and writing runbooks you hope you never need.

The scope is deliberately defensive. Monitoring Tailscale connectivity and fail2ban bans before they become incidents. Recovery runbooks for when a service wedges at 3 a.m. Tailscale is the only way in — WireGuard was removed to simplify the network stack. Port 22 is closed after Tailscale soaks. Validating backups actually restore, and surfacing the security decisions that were made because something needed to ship now.

Until now these tasks lived scattered — spin-offs from other projects, notes in a private repo, loose threads. Pulling them into their own project means they stop being somebody else's problem and start being work.

\n","linearSlug":"vps-hardening-30c7641caa7f"},{"slug":"other-agents","title":"other-agents","tagline":"the next robots after the coder","year":2026,"status":"completed","tags":["agents","linear","triage","coolify"],"cover":"/covers/other-agents.png","links":[{"label":"Linear project","url":"https://linear.app/gabistuff/project/other-agents-50b5c5b38cdd","kind":"linear","private":false}],"bodyHtml":"

When Linear Agents was one umbrella, it was holding too many ideas. The coder-agent earned its own project. What's left becomes this one — everything that isn't write code for me.

First candidate: triage-agent. A non-coding agent that watches the triage team, reads incoming issues, and classifies them against a cached taxonomy (project, priority, labels, likely assignee). Elicitation threshold so it asks me when confidence is low rather than guessing. Deployed at triage-agent.gabi.tv, same pattern as the coder-agent — Coolify application, Linear OAuth install, webhook on Issue.create.

After that, others arrive when the itch appears. A weekly-review agent. A research-assistant agent that takes an open question, runs it, and comes back with a document. A "factorise the shared scaffold" ticket sits here too — when three agents exist in production, the copy-paste pattern probably earns a shared library.

The rule stays the same as with the coder-agent: if the agent can do it end-to-end without me holding its hand, it stays. Otherwise it gets killed.

\n","linearSlug":"other-agents-50b5c5b38cdd"},{"slug":"storybridge-ai","title":"StoryBridge AI","tagline":"a writing tool escaping into open source","year":2026,"status":"dormant","tags":["oss","next.js","writing"],"cover":"/covers/storybridge-ai.png","links":[{"label":"storybridge-ai","url":"https://github.com/Coffelius/storybridge-ai","kind":"repo","private":false},{"label":"Linear project","url":"https://linear.app/gabistuff/project/storybridge-ai-oss-release-fc22e907c77c","kind":"linear","private":false}],"bodyHtml":"

StoryBridge AI started life as story-plotter-viewer — a Next.js 16 / React 19 app I built for myself to navigate sprawling plot trees without the cognitive tax of a generic outliner. It was private for a long time because it was weird, unfinished, and mine.

This project is the part where it stops being mine. Forking the local repo into a public one, swapping the default create-next-app README for something honest, adding the license, contributing guide, and code of conduct — the paperwork of letting other people touch your thing. MIT. No telemetry. No landing page trying to sell it to you.

If anyone else ends up using it, good. If nobody does, the act of making it releasable still clarifies what the tool actually is.

\n","linearSlug":"storybridge-ai-oss-release-fc22e907c77c"},{"slug":"storyplotter-mcp","title":"StoryPlotter MCP","tagline":"letting the chatbot read my fiction database","year":2026,"status":"dormant","tags":["mcp","writing","librechat","coolify"],"cover":"/covers/storyplotter-mcp.png","links":[{"label":"storyplotter-mcp","url":"https://github.com/Coffelius/storyplotter-mcp","kind":"repo","private":false},{"label":"Linear project","url":"https://linear.app/gabistuff/project/storyplotter-mcp-23f9dcef5900","kind":"linear","private":false},{"label":"LibreChat (host)","url":"https://chat.gabi.tv","kind":"other","private":false}],"bodyHtml":"

The plot database is a pile of JSON — characters, scenes, arcs, the backs of envelopes, the versions of the version. Useful to me. Opaque to any model that hasn't been handed it properly.

StoryPlotter MCP is a custom Model Context Protocol server that speaks this pile fluently: coherence checks across arcs, character roleplay with real continuity, scene generation that doesn't contradict chapter three, structured brainstorming with the database as ground truth. A sidecar on Coolify next to the chatbot, so every LibreChat conversation has the option of touching my actual story world.

It is the difference between an LLM that guesses and an LLM that remembers what you've already written.

\n","linearSlug":"storyplotter-mcp-23f9dcef5900"},{"slug":"fitbit-dump","title":"fitbit-dump","tagline":"reclaiming my own heartbeat from a company that rents it back to me","year":2026,"status":"dormant","tags":["oss","python","cli","data"],"cover":"/covers/fitbit-dump.png","links":[{"label":"fitbit-dump (private until OSS release)","url":"https://github.com/Coffelius/fitbit-dump","kind":"repo","private":true},{"label":"Linear project","url":"https://linear.app/gabistuff/project/fitbit-dump-open-source-prep-90f825265ca8","kind":"linear","private":false}],"bodyHtml":"

Fitbit has my sleep, my heart rate, my steps, my resting variability. They let me see it through an app. They do not really let me have it.

fitbit-dump is a small Python CLI that uses the official API to pull the raw JSON — all of it, incrementally — and drop it on my disk where it belongs. The original repo had accumulated too much: a phone companion, a Wear OS app, an HTML report with charts, phase tracking. For the open-source release I'm cutting it down to the bone: setup, sync, discover, status, reset, logout. Six commands. No dashboard. No analytics. Just the dump.

Strip the secrets, remove the personal data, write a real README, ship it. Someone else is going to want the same escape hatch.

\n","linearSlug":"fitbit-dump-open-source-prep-90f825265ca8"}]