CoStaff

01 Sound familiar?

The problems
CoStaff solves.

You don't need to understand how it works under the hood. You just need it to stop being a problem.

01 / 05

The frustration

You want to know "which product sold best last month" or "average order value from southern customers" — but you don't write SQL, so you queue up behind engineering. By the time you get the answer, the moment has passed.

With CoStaff

Just ask in plain English in your chat app. It explores your database, finds the right tables, and runs the query. You don't need to know what the schema looks like.

The frustration

A client emails an Excel file with columns out of order, mixed date formats, blank rows scattered through it. You spend 30 minutes cleaning it up by hand every single time — and next month, here comes another one.

With CoStaff

Drop the file in and say "clean this up, standard format." A tidy CSV comes back in seconds. It remembers the rules you used last time, so next month you don't even need to explain.

The frustration

Your boss wants a weekly report. You wrestle with Excel charts that still look ugly, the layout breaks the moment you paste into PowerPoint, and the final version is something you'd rather not look at.

With CoStaff

Send it the data. It picks the right chart type, writes the narrative, and ships a polished PDF or PPT with full multilingual font support — ready to hand over.

The frustration

Boss says "show me Q3 sales for the south region." You need a DBA to pull data, then build charts yourself, then write the narrative, then format the slides — three people, a week, just to get one report out.

With CoStaff

You just say "give me the Q3 south region sales report." Database agent queries → analyst agent formats. One chat window covers the whole pipeline, and what comes back is the PDF itself.

The frustration

You want one analysis: pull orders from Postgres, reconcile against an Excel a client emailed you, produce a multilingual chart-heavy report. That's at least 3 tools and 3 different roles.

With CoStaff

CoStaff dispatches internally: database agent pulls the orders → coding agent reconciles the Excel → analyst agent ships the report. One sentence in, one finished file out. You don't manage the handoffs.

02 How it works

Hand it off,
your AI Agent team takes it from there.

In your chat app, hand the work to Manager Agent — it parses intent and dispatches to the right specialists. You give the brief, you get the result.

Hand off

Just chat in Telegram, LINE, or WebChat — for example "Show me Q3 sales, make it a PDF." No new system, no new UI to learn.

Dispatch

Manager Agent parses your intent and routes work to the right specialist combination — DB Agent first if data lookup, then BA Agent for the report. Multi-agent handoff happens automatically, no manual wiring.

Specialists

Business, Coding, Database and other specialist Agents each work in their own domain, calling MCP tools as needed. Each Agent runs in its own container — independently scalable or replaceable.

Done

PDF, Excel, query results — back in your chat. You don't switch tools; everything happens in the chat window you were already using.

─── A complete conversation example ───

YOU

Show me Q3 sales, make it a PDF.

MANAGER

Got it. DB Agent first, then BA Agent.

DB AGENT

SQL query → 12,345 rows.

BA AGENT

Charts + narrative → final PDF.

YOU ◀

📎 Q3-report.pdf — Done, back in your chat.

View full user flow sequence diagram

03 System architecture

Six layers,
each independently replaceable.

CoStaff is not a single monolithic AI — it's a multi-agent + tool system stitched together. Each layer runs in its own container, does one thing, and can be swapped or extended independently.

USER

Hand off a task in a single chat message — you give the brief, you get the result.

CHANNELS

WebChat ・ Telegram ・ LINE ・ Discord ・ Slack (Docker container per channel)

MANAGER

Manager Agent — parses intent, remembers conversation, dispatches multi-agent work via A2A Protocol.

SPECIALISTS

Business ・ Coding ・ Database ・ Twinkle Hub ・ Custom Agent with MCP Server (each in its own container)

DATA

PostgreSQL + Shared Workspace (per-agent persistent storage)

LLM

Gemini ・ Gemma ・ Twinkle T1 ・ 3rd-party models (per-agent configurable)

View full technical architecture

04 Getting started

Just Docker.
Runs isolated in a container — clean and self-contained.

If you can run a Docker app, you can run CoStaff. From zero to handing work over to your AI staff: three commands, about 5 minutes.

Install the CLI

pip install costaff and you're done. Prerequisites are Python 3.10+ and Docker — both free, both already on most dev machines.

Run costaff onboard

It asks a few questions interactively: which AI model (Gemini / Claude / self-hosted)? Which chat app? Paste the corresponding bot token or API credential. Answers are written straight into .env — no config files to hand-edit.

Run costaff start

One command brings up every container: CoStaff Agent, MCP Server, your chosen chat-app channel. Hand over your first task in the chat app and your AI staff get to work — all running on your own machine.

05 One command for everything

Through the CoStaff CLI,
start, stop, and check service status quickly.

Commands are intuitively named and consistent — you know what they do at a glance, no docs lookup required.

$pip install costaff
✔ costaff installed
$costaff onboard
? Which AI model?  Gemini 2.5 Flash
? Telegram bot token?  ••••••••
✔ All set. Saved to .env
$costaff start
✔ Your AI agent is running
✔ Telegram bot is live
✔ Dashboard ready
$costaff dashboard
Opening http://localhost:18080 ...

You manage everything from the terminal — or from the web dashboard if you prefer clicking over typing.

costaff onboard	Interactive first-time setup
costaff start	Turn everything on
costaff stop	Turn everything off
costaff status	Check what's running
costaff logs	See what's happening
costaff dashboard	Open the web UI
costaff database backup	Save your data

07 Plans

CoStaff is an open-source AI staff platform

Free for personal use. For commercial inquiries, please fill out the form.
CoStaff source code is open, self-hostable, modifiable, and contributions are welcome. Personal use is free; for team or commercial use, please get in touch to learn more.

Open Source (OSS)

For personal use

Trial limits

Limited Agents
Limited Users
Limited Skills
Unlimited conversations

Get early access

Source will be public on GitHub at launch — PRs and issues welcome then.

Advanced Plan

Starter · Pro · Max · Enterprise

Everything in OSS, plus

Higher quotas: more agents, users, skills
Unlimited updates & upgrades during subscription
Built for teams and commercial environments
Pricing tailored to your scale

Get in touch to learn more

Community support: GitHub Issues / Discussions.
Tell me about your setup (team size, machines, use case) and you'll get a recommendation tailored to your needs.

08 Your data

It runs on your machine.
Not ours.

Most AI tools store your conversations on their servers and lock you into their model. CoStaff doesn't — your data lives 100% on your machine, and you decide whether AI inference goes offline too.

Your data stays on your machine

Conversations, reminders, tasks, files — all stored in a database on your own machine. Never synced to any cloud. We never see them.

AI inference can go fully offline (your choice)

Using Gemini / Claude / OpenAI? Prompts go to their API — but your conversation history still stays local.
Want 100% air-gapped? Plug in Ollama / vLLM / LM Studio, and even AI inference never leaves your network. The choice is yours.

Fully open source

At launch, the full source will be published on GitHub. Every step CoStaff takes with your data will be visible to you.

The problemsCoStaff solves.

Hand it off,your AI Agent team takes it from there.

Six layers,each independently replaceable.

Just Docker.Runs isolated in a container — clean and self-contained.

Through the CoStaff CLI,start, stop, and check service status quickly.

In the chat apps you use every day,your AI staff is ready and waiting.

CoStaff is an open-source AI staff platform

It runs on your machine.Not ours.

The problems
CoStaff solves.

Hand it off,
your AI Agent team takes it from there.

Six layers,
each independently replaceable.

Just Docker.
Runs isolated in a container — clean and self-contained.

Through the CoStaff CLI,
start, stop, and check service status quickly.

In the chat apps you use every day,
your AI staff is ready and waiting.

It runs on your machine.
Not ours.