For teams of 3–20 developers on Claude Code
Shared team memory for Claude Code
Your teammates' sessions know what each other are changing, what the team decided and what was assumed, before they touch the code. Everyone else follows the work, and answers the questions that are theirs, from a link.
Free during early access · then $9 per developer a month, free for one person (pricing) · never stores your code (what we keep)
Sound familiar?
The agents are fast. They just don't know about each other.
Worktrees keep sessions out of each other's files. Nothing keeps them out of each other's way. This is what changes.
- “Two of our sessions rebuilt the same page, two different ways.”Each session is told what teammates' sessions are changing in that part of the code, with their plan, before it starts.
- “Agents keep missing what we decided, and CLAUDE.md keeps growing.”Decisions are recorded as they're made and given to the sessions working where they apply. The newest call stands.
- “Sessions stop to ask, and the questions wait for me.”Agents state an assumption and keep going. Whoever knows confirms or corrects it from a link, and the agent hears at its next turn.
- “I can't tell why an agent did that, so review is the bottleneck.”Every change carries its reasons: what it's for, the calls made, what was assumed. The link goes in the pull request.
How it works
Context travels with the work, while you work.
Nobody writes anything down for anyone. It happens as your agents work, through Claude Code's hooks.
Before an agent touches the code
It's told what's changing in that area right now, the decisions that apply, what the business said about it, and what was already tried and dropped. Answer it yourself
When something is unclear
It states its assumption and keeps going. If the team's record has the answer, it comes straight back; if someone knows better, they correct it without stopping anyone. Turn unknowns into calls
When it builds on unfinished work
It builds against that work's plan. When the work lands, it's told exactly what changed from the plan. Build on what's coming
When it makes a call
The decision is recorded with its reason as it's made, checked against the ones that stand, and given to everyone the work touches. Decide by doing
When it stops
It leaves a handoff: what's done, what's left, the next step and what to watch out for. Whoever picks it up starts in minutes. Let the work speak
When a change is ready
It carries its reasons for the reviewer: what it's for, the calls made, what was assumed. Fix forward
What the team sees
One feed per project: what happened, and why it matters.
Assumptions with who should confirm them, decisions as they're made, work in progress with its plan, and what landed.
What Campfyrr caught for your team this week
- 🤔 Assumed, kept goingLee's agent assumed partial refunds don't return shipping, and kept going.Priya (product) is the one to confirm it.Refunds · 2 min agoAnswer it →
- ✅ DecidedAna decided amounts are stored in whole cents.Every session that touches billing is told; the older “decimals” call is marked replaced.Payments · 40 min ago
- 🔄 In progressBen is splitting checkout into three steps: details, payment, review.Two other sessions in checkout are building on his plan.Checkout · expected Thursday
- ✅ DonePriya's agent published the refund test results: all 42 pass, including partial refunds.Support can see exactly what customers will get.Refunds · 1 h agoOpen the results →
For the rest of the company
Founders, product and support see the work, and steer it, without reading code.
Talk wherever you already talk. Campfyrr is where everyone sees the same thing. Teammates who only read and answer are free.
- Where are we?One plain page per project: what's being worked on, what's waiting on whom, what landed. Send the link instead of a status meeting.
- Answer from a linkWhen an agent assumes something about pricing, a customer or a deadline, the person who knows confirms or corrects it in a click.
- Tell the agentsAdd a customer's words, a pricing call or a deadline to the part of the product it affects. The next agent working there is told first.
- Plans and results as pagesSessions publish a plan, test results or a sample output as a page with a link, written point first, for people who don't read code.
Set up
Two lines in your terminal, then it runs on its own.
Each developer installs the plugin and pastes their token from the connect page. Then invite your team.
claude plugin marketplace add sarkars/campfyrr claude plugin install campfyrr@campfyrr
- Nothing to fill inBriefs, assumptions, decisions, handoffs and reasons are recorded as your agents work.
- Private to your teamYour team's work stays inside the team, and Campfyrr never takes your code.
- Open hooksThe plugin is open source, so you can read every line before you install it. Read it
Before you install
What leads ask first
- How is this different from CLAUDE.md?
- CLAUDE.md is the same file for every session, and changes when someone edits it. Campfyrr is live: what teammates are changing right now, decided this morning or assumed last night, given to the session working in that part of the code. Keep CLAUDE.md for the stable rules.
- Doesn't Claude Code already share memory?
- Claude Code's projects, auto memory and cross-session messaging work across one person's sessions. Campfyrr shares across everyone on the team, and with the people around them.
- We already share sessions as links. Isn't that the same?
- A shared session is a transcript someone opens after the work is done. Campfyrr gives the next session the part that matters, at the moment it matters, without anyone sharing anything.
- Does it store our code?
- No. Hooks send file paths and what happened in the session, never file contents or diffs. The plugin's hooks are open to read, and the trust page lists exactly what's kept.
- What does it cost?
- Free during early access. After that, free for one person and $9 per developer a month for teams. Founders, product, sales and support who only read and answer are free.
- Does it only work with Claude Code?
- Claude Code gets the full plugin: hooks, commands and tools. Codex, Claude and any MCP client connect to the same memory with a token or by signing in.
VibeFlock: the open library
Seven principles. Every one removes a reason to stop.
The way of working Campfyrr is built on, written up in the open: 53 answers, 17 field guides and 17 glossary terms, free to read and to share.
- 1Keep moving
- 2Build on what's coming
- 3Fix forward
- 4Turn unknowns into calls
- 5Answer it yourself
- 6Let the work speak
- 7Decide by doing
- New · in two versionsASD-STE100 or Markdown for docs and agent instructions?What Simplified Technical English gives you, the 2026 buzz around it as an agent skill, and what it costs. Read it in ordinary English or in STE.
- AnswerStop being the human router for your team's agentsFour changes that take a lead out of the middle: assumptions, routing, recorded decisions and area briefs.
Field guides
VibeFlock in practice
Questions teams ask
Straight answers about Claude Code on a team
- How do I share memory across my team's Claude Code sessions?
- How do I stop Claude Code from stopping to ask questions?
- Why does our team's CLAUDE.md keep growing while agents still miss what we decided?
- How do I avoid two agents changing the same thing in different directions?
- How should I review a pull request an agent wrote?
- How do I show non-technical teammates what our Claude Code agents are building?
- Should we write docs and agent instructions in ASD-STE100 (Simplified Technical English) or Markdown?
- ASD-STE100 or Markdown? The same answer, written in Simplified Technical English
Give your team's sessions one memory.
Two minutes per developer. Free during early access.
claude plugin marketplace add sarkars/campfyrr claude plugin install campfyrr@campfyrrGet your token and install