One chat is a helper. Eight are a pile-up: the same files edited twice, decisions nobody can find, and a bill you only read afterwards. TEAM8 Build seats Claude Code chats as tech lead, backend, frontend, platform, security, QA and docs — each owning its own paths, claiming a file before it touches it, and writing what it did into your repository where your developers can read it.
Seating eight Claude chats is the easy part. These are the parts that stop them tripping over each other, and they are the reason TEAM8 exists.
The board, tickets, roadmap, decisions and file claims are Markdown and JSON committed next to your code. Every chat reads them, every developer reads them, git merges them — and they are still readable the day you stop using TEAM8. No database of ours holds your project.
Each role owns its paths. Before a chat touches anything outside them it claims the file, and a chat that reaches for a claimed file is stopped by a hook in Claude Code, not by a line in a prompt. Your developers are in the same record, so a chat is warned off what a person is already working on.
TEAM8 reads your own Claude plan and its live limits, then recommends a model and effort per role — Opus for the tech lead, Haiku for docs — and applies it in one click. Between tasks it compacts each chat back to its brief, claimed files and open tickets, so a long-running role stops dragging a full context window behind it.
Underneath, it is still a team of Claude chats you can watch work: replies, tool calls, checks and costs streaming in the window, with a person's approval required before anything merges or ships.
Every task gets a branch and a pull request. The role claims the files it touches, so frontend and backend never collide.
Lint, typecheck, tests and build must pass before a chat posts DONE. Another role reviews; security reviews anything touching sign-in, secrets or new dependencies.
Your release approver posts APPROVED before anything merges to main or reaches production. The release post carries the version, changelog and rollback plan.
Describe the site and Claude builds it in your repository while you watch it live. Then change it the way you'd point at a screen. It's real code in your project, not a locked template.
Site Check at phone widthEvery problem found on the live page, with a Fix button.
Next.js, Vite, Node APIs, React Native, Electron and Tauri, Django, Go, Rust, Kotlin, Swift, Flutter, .NET, Rails, Laravel, C++ and monorepos, each with its own checks.
GitHub, GitLab, CI, Docker, Kubernetes, Terraform, Prisma, Supabase, Vercel, Netlify, AWS, Sentry, .env files and more, each adding rules for Claude.
Every role owns its paths. Claims keep chats out of each other's files, and "who's working on what" warns your developers too.
Lint, typecheck, tests and build run from Runs & Outputs with logs and a clear pass or fail, before any DONE.
REVIEW, APPROVED and RELEASE posts on the board, architecture decision records and an incident procedure.
Threat model, auth and dependency reviews, secret handling and privacy checks, built into the flow.
Build and edit a website live: click to change, live theme, Site Check, screenshot design reviews and undo.
Test runs, builds, diffs and searches trimmed before Claude reads them, with the savings counted.
Pull and push from the title bar, patch notes on every push, and live "who's working on what" for your developers.
Vercel, Netlify, Cloudflare Pages, GitHub Pages, your own server or any deploy command, with one button.
Cost per day, model, role and session; your plan's limits; how full each chat's context is.
Deep royal blue and black, built to sit beside your editor. A standard window is one switch away.
From the current TEAM8 Build, running a demo Next.js CRM.
npm test, cargo build, a git diff of a refactor, a grep across src: every command a chat runs comes back as text it has to read, and most of it is noise. Token Saver runs those commands through RTK, which keeps the failures, errors and changed lines, and trims the rest before Claude reads them.
*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.
Illustration of the kind of trimming RTK does on a test run.
No. The team protocol requires an APPROVED post from a person before anything merges to main, releases or touches production, and TEAM8 chats can't answer permission prompts, so commands you haven't allowed are refused.
Yes. Open the folder: TEAM8 detects the stack and the tools you use, proposes checks that match your CI, and adds a team/ folder and rules to CLAUDE.md. Nothing already there is overwritten.
Real code in your project: plain HTML and CSS, Astro, Next.js, Vite + React, or your existing web stack. You own it, host it anywhere, and edit it with or without TEAM8.
Only the people who want the app. Everyone else sees normal branches, pull requests and plain team files in the repository.