“Is it still working, or waiting for me?”
Needs you shows every task that is blocked or waiting for a permission — and your phone tells you the moment it happens.
Open source · Self-hosted · Zero AI calls
SI is well,
I am well. All is well.
One calm page for what your AI coding agents are doing, what it costs you, and how you are doing — on your desk and on your phone.
SI = super intelligence: the AI agents working for you. Free and open source · self-hosted · for one person today, teams next
What is siiswell?
A dashboard for your AI agents — and for you.
siiswell is a free, open-source, self-hosted dashboard for AI coding agents. It shows what agents such as Claude Code, Codex and Gemini CLI are working on, which tasks are waiting for you, how many LLM tokens they use and what that would cost on the API — and it keeps an eye on the person in the middle with calm, rule-based well-being signals.
- SI
- Super intelligence: today's advanced AI and the agents built on it. “SI is well” means your agents are running smoothly.
- I am well
- You are not overloaded: no endless permission prompts, no buzzing phone at night, a long day made visible.
- Who it is for
- Developers and builders who run several AI agents at once. One person per installation today; team features are planned.
Works with the agents you already use
- Claude Code
- Codex
- Gemini CLI
- Claude Desktop
- Antigravity
- Grok
- opencode
- Manus
- Any MCP client
- Scripts & CI
Why siiswell
Agents never get tired.
People do.
AI agents now run for minutes or hours on their own, several at once, across terminals, desktop apps, servers and CI. Great for throughput. Hard on the person in the middle.
“My agents are everywhere.”
Laptop, server, desktop app, nightly script: one page for every machine, app and project.
“What did this month really cost?”
Token counts and the API-equivalent cost by day, hour, project and model — from your local logs.
“My phone buzzes all evening.”
Urgent things now, the rest in a summary. Quiet hours, Focus, and your calendar decide the rest.
“Long days creep up on me.”
A gentle You card: how busy today really was, a nudge to take a break, a wind-down note. No scores.
“Can I trust another dashboard?”
It never sees prompts, code or files, never calls an AI model, and you host it.
How it works
Agents report.
You look once.
- 1Agents report their long tasks. One short rule in an agent's instructions; it calls
report_taskwhen it starts andupdate_taskwhen it is blocked or done. Claude Code and Codex hooks add “waiting for you” by themselves. - 2Numbers, not content, are collected. A small collector turns local logs into hourly token totals. Your own activity is recorded only as “a prompt happened, at this time”.
- 3Fixed rules decide. Blocked, overdue, waiting, offline, near a plan limit: every minute, deterministic, explainable.
- 4You get the right push at the right time. On your desk, on your phone, or not at all during quiet hours.
Features
AI agent monitoring, LLM usage and well-being on one calm page.
Agent tasks
Needs you, Working now and Recent results. ETA, overdue markers, and stale tasks that tidy themselves up.
Usage & cost
A 12-month calendar, any day's 24 hours, and tokens and API-equivalent cost by host, app, project and model.
You, the person
Calm / Busy / Overloaded with the reasons, your usual range on every signal, gentle nudges you can switch off.
Calm notifications
Time-Sensitive now, a Scheduled Summary later. Quiet hours, Focus 60 min, and Focus from your calendar.
Privacy by default
No prompts, code or files. Project names hashed. Text scrubbed on your machine and again on the server.
Runs where you want
One Docker container, or a Cloudflare Worker. A zero-dependency CLI that updates itself from your dashboard.
Highlights
Your week in four plain sentences.
“About 12% less than the 7 days before.” “Every day ended before 23:00.” Big numbers, small units, one sparkline — in the spirit of the health apps you already know. Pin the cards you care about.
You
It looks after the person, not only the agents.
- Eight transparent signals, each with your usual range
- A pause after three hours, a note in your quiet hours
- Wind-down and good-morning messages; agents learn you are away and keep going
- No scores, no streaks, no rankings. Not a medical tool — and it says so
Usage & cost
LLM token usage and cost: a year at a glance, any hour in detail.
Claude Code and Codex logs become hourly totals per model — numbers only. See what your work would cost on the API, by machine, app, project and model, even on a flat-rate plan.
Privacy
A privacy line you can read.
siiswell records that you and your agents did something — never what.
Sent to your server
- Task status, ETA and a fixed kind (coding, docs…)
- A short task title, scrubbed and clipped (or none at all)
- Token counts per hour, model and app
- Time and type of your own interactions
- Project labels as a salted hash, unless you opt in
Never sent, never stored
- Prompts, replies, conversations
- Code, diffs, file contents, tool output
- Tool arguments and command text
- Working directories and local paths
- Real project names (unless you opt in)
Zero AINo model is ever called. It costs nothing to run, and every alert is a threshold you can read in the code.
Yours to hostSingle-tenant by design. Your data stays in your container or your Cloudflare account.
One token per appEach machine and app has its own key; a lost one is revoked alone.
Get started
Running in about ten minutes.
Recommended
Docker
A home server, a VPS or a NAS. Built-in sign-in with one password: set PUBLIC_URL and ADMIN_PASSWORD in .env before you start.
git clone https://github.com/jcaol/siiswell.git siiswell && cd siiswell
cp .env.example .env
docker compose up -d --buildFor Workers users
Cloudflare
A Worker and D1 in your own account, protected by Cloudflare Access.
bunx wrangler d1 create siiswell
bunx wrangler d1 migrations apply siiswell --remote
bun run deployOn each computer
Connect your agents
Installs the siiswell command, adds it to your agent apps, and can add the usage collector.
bun install && bun run build:cli
scripts/setup-client.sh
siiswell report "Hello, siiswell" --kind testingThe full guide — HTTPS, phone pushes, tokens, backups — is in the README on GitHub, in eight languages.
Open source
Open, and fair to everyone who builds on it.
The server is AGPL-3.0, so improvements to a hosted siiswell stay open. The client is Apache-2.0, so it installs on any machine — including company laptops — without licence questions. A commercial licence is available for the server — ask us. Source code: github.com/jcaol/siiswell.
| Part | Licence |
|---|---|
| Server and web page | AGPL-3.0-only |
| CLI, bridge, shared client code | Apache-2.0 |
| Contributions | Contributor License Agreement |
| Commercial use without AGPL | Commercial licence |
FAQ
Questions people ask.
What does “SI is well, I am well” mean?
SI stands for super intelligence: the AI agents working for you. “SI is well” means your agents are running smoothly; “I am well” means you are not overloaded. siiswell watches both, so all is well.
How do I know when Claude Code is waiting for my permission?
siiswell's Claude Code hooks mark a session as “waiting for you” the moment it asks for a permission, list it under Needs you, and send a push to your phone — urgent items at once, everything else in a summary.
How do I track Claude Code and Codex token usage and cost?
The siiswell usage collector reads the token counts in your local Claude Code and Codex logs (never their content) and shows tokens and API-equivalent cost by day, hour, machine, app, project and model.
Can working with AI agents lead to burnout?
Supervising several agents means constant context switching and quick approvals, and many developers report feeling drained by it. siiswell makes the load visible — active hours, quiet-hour work, response pace — batches notifications, and suggests breaks. It is not a medical tool.
Do my agents need special support?
No. Any agent that can use MCP tools can report tasks: add the server and a short rule to its instructions. Without MCP, the siiswell command or a plain HTTP call works too.
Does it work with cloud agents like Manus, Grok or ChatGPT?
For task reporting, yes wherever the agent can add a remote MCP server with a bearer token. Usage and the “You” signals come from local logs and hooks, so they cover the agents on your own machines.
Does it slow my agents down or cost tokens?
Barely. A report is a few hundred bytes, only for tasks longer than about ten minutes. Hooks return at once. siiswell itself never spends AI tokens.
How accurate is the cost?
It is the API-equivalent price: token counts times a per-model price table you can override. On a subscription it shows what the same work would cost on the API, not your bill.
Is the “You” card a health tool?
No. It counts timestamps and events with fixed thresholds and says so on the page. It makes a long day visible; it does not judge it. Every nudge can be switched off.
I run an agent orchestrator. Will its agents count as me?
Not if you mark its projects with a .siiswell-agents file (or start its agents with SIISWELL_AGENT=1). Only your own sessions count, and nothing is reported twice.
Is siiswell for teams?
Today siiswell is built for one person: one installation, one owner. Team features — shared views of agents and LLM costs, with personal well-being data kept private to each person — are on the roadmap.
What does it cost to run?
Nothing extra: one small container and a SQLite file, or a personal Cloudflare installation that stays well inside the free D1 limits.