01
Who owns what?
Every role has its own files and folders, and a chat claims a file before touching someone else's. No two chats silently editing the same thing.
TEAM8 · for Windows 10 & 11
Your project.
Your project.
A team of humans
and Claude.
TEAM8 seats Claude Code chats in roles next to the people on your project. Everyone works from one board, with a plan, reviews, approvals and costs in one window. One download, three editions: start free, and a key unlocks the one made for what you're building.
Free edition, no accountKeys are one-time3 computers per keyRuns on your Claude plan
Three editions
Made for what
Made for what
you make
Start here
For apps and software
Build
Tech lead, backend, frontend, platform, security, QA and docs, each owning its own paths. 20 stacks recognised, nothing merges without a person's approval, and an AI Site Builder.
For games
Free
Gaming
Lead, combat, animator, netcode, fx, qa and more. 66 engines recognised, git co-op with your team and one-button deploys to itch.io and the web.
For every other project
Basic
The download, as it comes. Operations, research, writing, marketing, design, finance and review, working with your documents, your email and your calendar.
1download, with all three editions in it
8preset roles in each edition, plus your own
0API keys: it runs on your own Claude plan
0to start: Basic is free, keys are one-time
Why TEAM8
One Claude chat is a helper.
One Claude chat is a helper.
Five are a mess.
02
What's actually done?
One board, tickets, a roadmap and a definition of done, saved as plain files in your project. Chats post START, DONE and BLOCKED; people post decisions and approvals.
03
What did it cost?
TEAM8 reads Claude Code's own records: cost per day, model, role and session, your plan's limits, and the tokens Token Saver kept out of your chats.
01
Open your project
TEAM8 recognises what's in the folder (a game engine, a software stack, or your documents and apps) and sets up the team: board, roles, tickets, roadmap and rules for Claude.
02
Seat the roles
One click seats a Claude chat in a role. It reads its brief, checks its tools and posts READY. Your people join the same board from their own TEAM8.
03
Run it together
Chats do the work and report on the board. People review, approve and decide. You steer from one dashboard and see what every role costs.
In every edition
The same core.
The same core.
Different teams.
01
Claude, inside the app
Uses your own Claude Code. Seat a role in one click; replies, tool calls and costs stream in live.
02
Roles & ownership
Eight preset roles per edition, or your own. Each owns its files, and claims keep chats out of each other's way.
03
Board, tickets, roadmap
A team board, a tracker and milestones, stored as Markdown and JSON inside your project.
04
People approve
Reviews, approvals and decisions are posts on the board, so what ships, sends or merges is a person's call.
05
Decisions with defaults
A chat that hits a choice it should not make raises it, with what happens if you say nothing, and carries on under that default instead of guessing or stalling.
06
Usage, limits, prompts
Cost per day, model, role and session; your 5-hour and weekly limits; every prompt, searchable.
07
Token Saver
Command output trimmed before Claude reads it, with every saved token counted per session and project.
08
Terminal & files
A real terminal and a file browser with previews, without leaving the window.
09
Plugin included
The same team commands work in any Claude Code chat: the terminal, your editor or the Claude desktop app.
10
Its own window
A shaped window per edition (hexagon, tall box or capsule) with a standard window one switch away.
Claude already has agents
Subagents forget.
Subagents forget.
A team remembers.
Claude Code can already fan a job out to subagents, and its agent teams can run several instances at once. They are built for parallel coding: each one starts cold, does its task, reports back and is gone. That is the right tool for "debug these three things now". It is not a team, and TEAM8 does not replace it — every TEAM8 role is a full Claude Code chat, so it can still fan out to subagents inside its own task.
01
Cold start, or a role that knows its job
A subagent gathers its context from scratch every time, and pays for that before it does any work. A TEAM8 role is a chat that stays: its brief, the files it owns, its open tickets and the team's record are there to re-read instead of rediscover.
02
A summary, or a record
When a subagent finishes, what survives is a summary inside someone else's context. In TEAM8 what survives is written into your project: the board post, the ticket, the claim, the decision. The next chat reads it, and so does the next person, months later.
03
Self-coordinating, or accountable
Agent teams decide among themselves who does what. TEAM8 puts a person at the gate: roles own their files, claim before they touch anyone else's, and nothing merges, sends or ships without someone saying so on the board.
The bill is context, not the model
The team TEAM8 grew from measured its own first week across seven chats: cache writes were 59% of the estimated cost, and the average context when a chat woke up was 627K tokens. Which model each role runs was the last lever, not the first. That is one team's week, not a law — but it is why TEAM8 measures the same things for you.
- Compaction between tasks. When a role finishes a task, its chat is compacted down to its brief, the files it claimed and its open tickets, so the next task doesn't drag the last one behind it.
- Cost per role, per day, per session. Read from Claude Code's own records, beside your plan's limits and how full each chat's context is.
- Model and effort per role. Advised from what your plan can actually afford, applied in one click.
- Token Saver. Command output trimmed before Claude reads it, with every saved token counted.
Token Saver · in every edition
Same work.
Same work.
Fewer tokens.
Every command a chat runs (a test run, a build, a git diff, a search) comes back as text Claude has to read, and most of it is noise. Token Saver runs those commands through RTK, which keeps the lines that matter and trims the rest before Claude reads a word.
60–90%less output on common
dev commands*
dev commands*
1switch in
Settings
Settings
0changes to how
you work
you work
- Limits go further. Less noise in each chat's context means more work before the 5-hour and weekly limits, and emptier context windows.
- Every saved token, counted. Token Usage shows what was saved in each session, each project and all time, which commands save the most, and what it's worth.
- Local and open source. RTK runs on your computer and sends nothing anywhere. TEAM8 finds it, or installs it with one command.
*RTK's published figure for common commands such as git, test runners, builds and search. Savings depend on what your chats run. Token Saver is powered by RTK (Apache-2.0), an independent open-source project not made by TEAM8.
A test run, as Claude reads it
without npm test 412 lines ≈ 6,100 tokens
with rtk npm test 3 lines ≈ 70 tokens
FAIL src/api/orders.test.ts › rejects an empty cart
expected 400, received 500 (orders.ts:88)
1 failed · 127 passed · 4.2 s
Illustration of the kind of trimming RTK does on a test run.
Yours, not ours
No account. No cloud.
No account. No cloud.
No lock-in.
Your Claude plan
TEAM8 drives the Claude Code already on your computer. No API key to paste, and no markup on your AI usage.
Plain files
The board, tickets and roadmap are Markdown and JSON in your project. Stop using TEAM8 and you keep every word.
No TEAM8 servers
No sign-up with us and no tracking in the app. Your work goes only where you send it.
Free, then pay once
Basic is free and whole. A key for Build or Gaming is one payment, no subscription, and it keeps working. Included: Free updates through version 1.x.
Which edition is for me?
Building an app, a website with code, an API or any other software: Build. Making a game: Gaming. Running anything else with several roles, like a business, a launch, an event or a report: Basic. Compare them side by side.
How is this different from Claude's own subagents?
Subagents and agent teams are built into Claude Code for parallel work: each one starts with an empty context, does a task and reports a summary back. TEAM8 keeps a team instead — roles that stay, own their files, and write what they did into your project so the next chat and the next person can read it. The two work together: a TEAM8 role is a full Claude Code chat, so it can still fan out to subagents inside its own task. The longer answer.
Do I need a Claude subscription?
Yes. TEAM8 works through Claude Code on your computer, so you need Claude Code signed in with a Claude plan that includes it (such as Pro or Max) or with API billing. TEAM8 doesn't include or resell Claude usage.
Is Basic really free?
Yes. Download TEAM8 and Basic is what you get: roles, board, tickets, roadmap, research, terminal, files, token usage, plan limits, Token Saver, the Claude Code plugin, and Communications and Social. No account with us, no trial clock, nothing switched off after a fortnight.
What does a key change?
A key turns the same download into Build or Gaming: that edition's roles and protocol, what it recognises in a project (20 software stacks or 66 game engines, each with its checks), and the tools built on them — Site Builder, Rig Builder, git co-op, patch notes and deploys. You redeem it in the app under Editions, and you can switch between the editions you own whenever you like.
Is a key really a one-time payment?
Yes. Pay once and the key keeps working on up to 3 of your computers, with no subscription. Included: Free updates through version 1.x.
Does each teammate need one?
Everyone can download TEAM8 free. A teammate needs their own key only for Build or Gaming. The files TEAM8 writes are plain text, so anyone can read them with any edition.