# ArcRoom AI

> ArcRoom is a team playbook for your AI agents, so they never redo work that is already done. Answers, the steps that worked, failed attempts, decisions (with why; a newer one retires the older) and rules live in one place. Every agent reads it before it works and writes to it when it finishes; if the work was already done, by you, a teammate or another agent, it is reused. A big job is split into parts, one level at a time (at most 8 parts, 5 levels deep); a part any job can use is kept as a reusable workflow, and agents open a part by its id only when they start it. Your notes stay yours; finished work is shared with your team, and you can keep anything private. Any agent can use it: it works with the agents you already use (see Agent support below), any other MCP agent uses its tools, and your own agents use the HTTP API. Once set up, each prompt (first 1,000 characters) and the commands and file paths of each tool call are recorded, while file contents, tool output and diffs stay on the laptop unless the agent saves them to memory. Any agent, any work. Free for 2 people; Pro is $8 a person a month.

## Docs

- [Quickstart](https://www.arcroomai.com/docs): make a key, run `npx arcroom init`, add other agents with `--agent codex|cursor|opencode`
- [Agent support](https://www.arcroomai.com/docs/agents): what works in Claude Code, Codex, Cursor, OpenCode (experimental), any MCP agent and the HTTP API
- [How recall works](https://www.arcroomai.com/docs/recall): on each prompt ArcRoom hands over answers, steps, decisions and the code map
- [What is sent](https://www.arcroomai.com/docs/data): every field that leaves the laptop, and what never does
- [Isolation and controls](https://www.arcroomai.com/docs/isolation): workspaces, what stays yours and what is shared with your team, `arcroom off`, revoking a key, deleting a workspace
- [Fail-open behaviour](https://www.arcroomai.com/docs/fail-open): if ArcRoom fails, your agent carries on without it; recall waits at most 3 seconds
- [Done-checks](https://www.arcroomai.com/docs/done-checks): `.arcroom.json`, `arcroom checks`, approval by code
- [Rules, answers, steps and expiry](https://www.arcroomai.com/docs/memory): the rules every agent follows, what is stored, how long answers are reused, how savings are estimated
- [Troubleshooting](https://www.arcroomai.com/docs/troubleshooting)
- [Any MCP agent](https://www.arcroomai.com/docs/mcp): ArcRoom's tools on their own, after `arcroom init`
- [HTTP API for your own agents](https://www.arcroomai.com/docs/api): find, claim, complete over HTTP, with Python and TypeScript examples
- [Words you'll see](https://www.arcroomai.com/docs/faq): reused, replay, in flight, hand-over, done-checks, tokens saved
- [ArcRoom vs Claude Code memory](https://www.arcroomai.com/docs/compare/claude-code-memory)
- [ArcRoom vs other agent memory tools](https://www.arcroomai.com/docs/compare/agent-memory-tools): agentmemory, Memorable, Mem0 (OpenMemory), ByteRover

## Trust

- [Security](https://www.arcroomai.com/security): data flow, retention, how your data is protected, controls, disclosure
- [Privacy policy](https://www.arcroomai.com/privacy)
- [Terms of use (beta)](https://www.arcroomai.com/terms)
- Contact (support, privacy and security reports): sachin.kumar@arcroomai.com
- [Pricing](https://www.arcroomai.com/pricing): Free (2 people, 1 workspace, 1 project, 1,000 memories, 7 days of history), Pro at $8 a person a month or $6 billed yearly (25,000 memories a person, 90 days of history), Enterprise on request.
- [Status](https://www.arcroomai.com/status)
- [Changelog](https://www.arcroomai.com/changelog)
- [Blog](https://www.arcroomai.com/blog): [Your agent already did this work once](https://www.arcroomai.com/blog/your-agent-already-did-this-work-once)

## Use cases

Example scenarios that work today, each with the steps, what the agent sees, and how it compares with other tools.

### Solo builders

- [Last night's work, not redone tonight](https://www.arcroomai.com/usecases/starting-tomorrow-where-you-stopped-tonight): Solo founder, one product, evenings and weekends
- [The question you ask every week](https://www.arcroomai.com/usecases/the-question-you-ask-every-week): Solo developer on a codebase older than a month
- [Two terminals, one repo, no collisions](https://www.arcroomai.com/usecases/two-terminals-one-repo-no-collisions): Solo builder running two agents side by side
- [The agent that keeps running the broken command](https://www.arcroomai.com/usecases/the-agent-that-keeps-running-the-broken-command): Anyone who leaves an agent running
- ["Done" means the checks pass](https://www.arcroomai.com/usecases/done-means-the-checks-pass): Solo developer tired of 'done' that isn't
- [Decisions you made weeks ago](https://www.arcroomai.com/usecases/decisions-you-made-weeks-ago): Solo founder who changed their mind once and doesn't want to again
- [Stop re-exploring the same folders](https://www.arcroomai.com/usecases/stop-re-exploring-the-same-folders): Solo developer on a large repo
- [Don't try the dead end twice](https://www.arcroomai.com/usecases/don-t-try-the-dead-end-twice): Solo builder who already learned the hard way
- [Two months later, nothing figured out twice](https://www.arcroomai.com/usecases/coming-back-after-two-months-away): Solo founder returning to a paused project
- [One backend, a mobile app and a web app](https://www.arcroomai.com/usecases/one-backend-a-mobile-app-and-a-web-app): Indie hacker shipping an iOS app and a web app on one API
- [The same bug, reported by five users](https://www.arcroomai.com/usecases/the-same-bug-reported-by-five-users): Solo founder answering support email between features
- [An App Store rejection, remembered](https://www.arcroomai.com/usecases/an-app-store-rejection-remembered): Solo iOS developer who has been through App Review a few times
- [Pricing experiments, with the results kept](https://www.arcroomai.com/usecases/pricing-experiments-with-the-results-kept): Indie hacker testing prices on a small SaaS
- [A game jam where no broken fix is tried twice](https://www.arcroomai.com/usecases/a-weekend-game-jam-alone): Hobby game developer doing a 48-hour jam in Godot
- [A cofounder who writes the copy, with an agent](https://www.arcroomai.com/usecases/a-cofounder-who-writes-the-copy-with-an-agent): Two-person startup: one developer, one non-technical cofounder
- [A library author and the same issue, again](https://www.arcroomai.com/usecases/an-open-source-author-and-the-same-issue-again): Maintainer of a small TypeScript library, alone
- [Learning Rust after years of TypeScript](https://www.arcroomai.com/usecases/learning-rust-after-years-of-typescript): Experienced web developer learning a new language

### Small teams

- [Your teammate's agent already found the answer](https://www.arcroomai.com/usecases/your-teammate-s-agent-already-found-the-answer): Team of 4 on one SaaS codebase
- [Two developers, same bug, same morning](https://www.arcroomai.com/usecases/two-developers-same-bug-same-morning): Team with a shared bug queue
- [A new hire's first week](https://www.arcroomai.com/usecases/a-new-hire-s-first-week): Startup adding its 6th engineer
- [Handing a half-done branch to a teammate](https://www.arcroomai.com/usecases/handing-a-half-done-branch-to-a-teammate): Team where someone goes on leave
- [A wrong memory, caught](https://www.arcroomai.com/usecases/a-wrong-memory-caught): Team that trusts its memory
- [The answer that went stale](https://www.arcroomai.com/usecases/the-answer-that-went-stale): Team with fast-moving code
- [Decisions with the why attached](https://www.arcroomai.com/usecases/decisions-with-the-why-attached): Team that keeps asking 'why did we do this?'
- [One definition of done for everyone](https://www.arcroomai.com/usecases/one-definition-of-done-for-everyone): Team with mixed agents and habits
- [Monorepo, only the memory that matters](https://www.arcroomai.com/usecases/monorepo-only-the-memory-that-matters): Frontend and backend developers sharing one repo
- [What did our agents do this week?](https://www.arcroomai.com/usecases/what-did-our-agents-do-this-week): Tech lead of a 5-person team
- [From the designer's decisions to the coder's agent](https://www.arcroomai.com/usecases/from-the-designer-s-decisions-to-the-coder-s-agent): Product team with one designer and three frontend developers
- [Support engineer and developer, one investigation](https://www.arcroomai.com/usecases/support-engineer-and-developer-one-investigation): SaaS team with one support engineer and four developers
- [Pairing across two laptops](https://www.arcroomai.com/usecases/pairing-across-two-laptops): Two developers pairing on one feature, each with an agent
- [Release captain, a new one every week](https://www.arcroomai.com/usecases/release-captain-a-new-one-every-week): Team of eight with a rotating release captain
- [Two people, two decisions that clash](https://www.arcroomai.com/usecases/two-people-two-decisions-that-clash): Team where two leads own overlapping areas
- [An intern's agent that learns from review](https://www.arcroomai.com/usecases/an-intern-s-agent-that-learns-from-review): Engineering team with a summer intern and a mentor
- [Clearing a code review backlog](https://www.arcroomai.com/usecases/clearing-a-code-review-backlog): Team of five with 20+ open pull requests
- [Private first, shared when it's ready](https://www.arcroomai.com/usecases/private-first-shared-when-it-s-ready): Developer on a team of five, trying ideas before sharing them

### Many agents, one playbook

- [Cursor for UI, Claude Code for the backend](https://www.arcroomai.com/usecases/cursor-for-ui-claude-code-for-the-backend): Full-stack builder who likes different tools for different work
- [Trying a new agent without losing your history](https://www.arcroomai.com/usecases/trying-a-new-agent-without-losing-your-history): Developer testing OpenCode
- [One agent writes, another reviews](https://www.arcroomai.com/usecases/one-agent-writes-another-reviews): Developer who uses Codex as a second pair of eyes
- [The day's work, not redone at night](https://www.arcroomai.com/usecases/laptop-by-day-desktop-by-night): Solo builder with two machines
- [Four worktrees, four agents](https://www.arcroomai.com/usecases/four-worktrees-four-agents): Developer running agents in parallel git worktrees
- [A teammate on a different agent](https://www.arcroomai.com/usecases/a-teammate-on-a-different-agent): Two developers who each prefer their own agent
- [A fix started from the phone, not redone at the laptop](https://www.arcroomai.com/usecases/started-on-the-phone-finished-at-the-laptop): Developer who starts tasks from the phone while away
- [The work agent at work, your own agent at home](https://www.arcroomai.com/usecases/the-work-agent-at-work-your-own-agent-at-home): Developer with an employer-approved agent and personal side projects

### Research and writing

- [Competitor research done once, used by everyone](https://www.arcroomai.com/usecases/competitor-research-done-once-used-by-everyone): Founder and a small team
- [Writing a pitch across ten sessions](https://www.arcroomai.com/usecases/writing-a-pitch-across-ten-sessions): Founder writing accelerator applications
- [Market sizing with the assumptions kept](https://www.arcroomai.com/usecases/market-sizing-with-the-assumptions-kept): Founder or analyst
- [Customer interview notes, many sessions](https://www.arcroomai.com/usecases/customer-interview-notes-many-sessions): Founder doing discovery calls
- [The weekly report that rebuilds itself](https://www.arcroomai.com/usecases/the-weekly-report-that-rebuilds-itself): Ops or growth person using an agent every Monday
- [House style that sticks](https://www.arcroomai.com/usecases/house-style-that-sticks): Writer or marketer using agents for content
- [Data cleaning steps you won't redo](https://www.arcroomai.com/usecases/data-cleaning-steps-you-won-t-redo): Analyst working on CSVs with an agent
- [Same answers, many application forms](https://www.arcroomai.com/usecases/same-answers-many-application-forms): Founder applying to several accelerators
- [A glossary for every language](https://www.arcroomai.com/usecases/a-glossary-for-every-language): Team translating a product into Hindi and Spanish
- [Planning a launch over two weeks](https://www.arcroomai.com/usecases/planning-a-launch-over-two-weeks): Founder running a launch from the terminal

### Sales

- [Account research one rep did, reused by the next](https://www.arcroomai.com/usecases/account-research-one-rep-did-reused-by-the-next): B2B sales team doing account research in Claude Code
- [Proposals built from approved sections](https://www.arcroomai.com/usecases/proposals-built-from-approved-sections): Sales engineers writing proposals in a git repo of Markdown templates
- [RFP answers reused, but only while they're true](https://www.arcroomai.com/usecases/rfp-answers-reused-but-only-while-they-re-true): Mid-size SaaS team answering security questionnaires and RFPs with Claude Code
- [Call prep that builds on the last call](https://www.arcroomai.com/usecases/call-prep-that-builds-on-the-last-call): Account executive running a five-call enterprise deal
- [Battlecards that know their age](https://www.arcroomai.com/usecases/battlecards-that-know-their-age): Product marketing and sales reps using agents for competitive deals

### Marketing and growth

- [SEO posts that don't fight over one keyword](https://www.arcroomai.com/usecases/seo-posts-that-don-t-fight-over-one-keyword): Content team writing blog posts in an MDX repo with Claude Code and Codex
- [Ad copy where the losing angles stay lost](https://www.arcroomai.com/usecases/ad-copy-where-the-losing-angles-stay-lost): Performance marketer writing ad variants in Codex
- [The second launch learns from the first](https://www.arcroomai.com/usecases/the-second-launch-learns-from-the-first): Startup growth team running launches from Claude Code
- [Analytics questions answered once](https://www.arcroomai.com/usecases/analytics-questions-answered-once): Growth team asking Claude Code questions over a dbt repo
- [One brand voice in three different agents](https://www.arcroomai.com/usecases/one-brand-voice-in-three-different-agents): Marketing team writing in Claude Code, Codex and Cursor

### Product and design

- [A PRD written over two weeks, decisions kept](https://www.arcroomai.com/usecases/a-prd-written-over-two-weeks-decisions-kept): Product manager writing PRDs in the repo's docs folder with Claude Code
- [Usability findings tied to the screen they're about](https://www.arcroomai.com/usecases/usability-findings-tied-to-the-screen-they-re-about): Product designer and researcher running usability tests
- [Roadmap 'no's that stay no](https://www.arcroomai.com/usecases/roadmap-no-s-that-stay-no): Product team deciding what not to build
- [Design-system rules Cursor actually follows](https://www.arcroomai.com/usecases/design-system-rules-cursor-actually-follows): Design engineer and frontend team using Cursor and Claude Code
- [From spec to tickets without filing twice](https://www.arcroomai.com/usecases/from-spec-to-tickets-without-filing-twice): PM and two engineers turning a spec into GitHub issues with agents
- [Why is this feature flag still on?](https://www.arcroomai.com/usecases/why-is-this-feature-flag-still-on): Product and engineering team with flags in code
- [Release notes in the customer's words](https://www.arcroomai.com/usecases/release-notes-in-the-customer-s-words): PM writing a monthly customer changelog with Claude Code
- [PM and engineer agents in one room](https://www.arcroomai.com/usecases/pm-and-engineer-agents-in-one-room): PM in Claude Code and engineers in Codex, same repo

### Long-running agents

- [Bot-driven sessions that don't redo the last one's work](https://www.arcroomai.com/usecases/a-telegram-bot-that-drives-claude-code-at-home): Developer who runs claude -p from a chat bot on a home PC
- [A three-day migration where no part is done twice](https://www.arcroomai.com/usecases/a-three-day-job-through-restarts-and-compaction): Developer leaving Claude Code on a long migration over a weekend
- [Three scheduled agents, nothing done twice](https://www.arcroomai.com/usecases/three-agents-in-a-row-while-you-sleep): Solo founder scheduling headless agents on a home machine
- [Coding agents in containers, one per issue](https://www.arcroomai.com/usecases/coding-agents-in-containers-one-per-issue): Team running Claude Code headless in containers on its own server

### Building agents

- [An agent that learns from its own failed attempts](https://www.arcroomai.com/usecases/an-agent-that-learns-from-its-own-failed-attempts): Builder whose agent runs on Claude Code headless (claude -p)
- [An eval loop that reads its own past failures](https://www.arcroomai.com/usecases/an-eval-loop-that-reads-its-own-past-failures): Team improving an agent against a 120-task eval set
- [A prompt that gets better each week](https://www.arcroomai.com/usecases/a-prompt-that-gets-better-each-week): Builder tuning the system prompt of a support-reply agent

### CI and automation

- [Laptop work not redone by the CI agent](https://www.arcroomai.com/usecases/started-on-the-laptop-finished-by-the-ci-agent): Small team with Claude Code running headless in GitHub Actions
- [Done-checks in CI without a terminal](https://www.arcroomai.com/usecases/done-checks-in-ci-without-a-terminal): Team that wants the same 'done' in CI as on laptops
- [A nightly dependency upgrader that remembers](https://www.arcroomai.com/usecases/a-nightly-dependency-upgrader-that-remembers): Team running an agent every night to bump packages
- [A PR review bot that knows your decisions](https://www.arcroomai.com/usecases/a-pr-review-bot-that-knows-your-decisions): Team with an agent reviewing every pull request
- [Flaky tests, known as flaky](https://www.arcroomai.com/usecases/flaky-tests-known-as-flaky): Team with a slow, noisy test suite
- [A changelog that doesn't start from zero](https://www.arcroomai.com/usecases/a-changelog-that-doesn-t-start-from-zero): Team generating release notes with an agent
- [An agent woken by an alert](https://www.arcroomai.com/usecases/an-agent-woken-by-an-alert): Team with an on-call agent triggered by errors
- [Runaway spend in an unattended job](https://www.arcroomai.com/usecases/runaway-spend-in-an-unattended-job): Anyone running agents with nobody watching
- [A matrix of agents, split without overlap](https://www.arcroomai.com/usecases/a-matrix-of-agents-split-without-overlap): Team splitting a large refactor across CI jobs
- [Nightly docs that know the code map](https://www.arcroomai.com/usecases/nightly-docs-that-know-the-code-map): Team keeping docs in sync with code

### DevOps, SRE and security

- [Terraform changes with the decision attached](https://www.arcroomai.com/usecases/terraform-changes-with-the-decision-attached): Platform team changing infrastructure with Claude Code
- [Kubernetes debugging, the second time](https://www.arcroomai.com/usecases/kubernetes-debugging-the-second-time): SRE debugging a cluster with an agent in the terminal
- [Runbooks the agent actually follows](https://www.arcroomai.com/usecases/runbooks-the-agent-actually-follows): Ops team with runbooks in the repo
- [Postmortems turned into memory](https://www.arcroomai.com/usecases/postmortems-turned-into-memory): Engineering team writing blameless postmortems
- [A cost investigation done once, not redone by a teammate](https://www.arcroomai.com/usecases/a-cloud-bill-investigation-picked-up-next-week): Startup CTO and an engineer chasing an AWS cost spike
- [Security review findings that stay found](https://www.arcroomai.com/usecases/security-review-findings-that-stay-found): Team running authorized security reviews of its own code
- [Vulnerability triage that isn't repeated](https://www.arcroomai.com/usecases/vulnerability-triage-that-isn-t-repeated): Team triaging dependency advisories from a scanner
- [Compliance evidence, gathered once a quarter](https://www.arcroomai.com/usecases/compliance-evidence-gathered-once-a-quarter): Small company preparing for its own audit
- [On-call handover between shifts](https://www.arcroomai.com/usecases/on-call-handover-between-shifts): On-call rotation across two shifts
- [An infra repo full of secrets](https://www.arcroomai.com/usecases/an-infra-repo-full-of-secrets): Ops team with kubeconfigs, tfvars and vault files in reach

### Data, ML and research

- [Which settings already failed](https://www.arcroomai.com/usecases/which-settings-already-failed): ML engineer running training experiments through Claude Code
- [Fine-tuning a small model without repeating bad runs](https://www.arcroomai.com/usecases/fine-tuning-a-small-model-without-repeating-bad-runs): Small team fine-tuning a small local model
- [A notebook analysis where no step is redone](https://www.arcroomai.com/usecases/a-notebook-analysis-across-four-sessions): Data scientist working in Jupyter with an agent
- [dbt questions answered once](https://www.arcroomai.com/usecases/dbt-questions-answered-once): Analytics engineers working in a dbt project with agents
- [SQL answers analysts reuse](https://www.arcroomai.com/usecases/sql-answers-analysts-reuse): Analysts writing warehouse queries with Claude Code
- [Reading a research paper once for the whole lab](https://www.arcroomai.com/usecases/reading-a-research-paper-once-for-the-whole-lab): Research group using agents to read and compare papers
- [Dataset quirks every agent should know](https://www.arcroomai.com/usecases/dataset-quirks-every-agent-should-know): Data team with an old event table full of surprises
- [One definition of 'active user'](https://www.arcroomai.com/usecases/one-definition-of-active-user): Product analytics team after a metric argument

### Internal agents

- [Babysitting data pipelines](https://www.arcroomai.com/usecases/babysitting-data-pipelines): Data team with an agent watching nightly jobs
- [Meetings that end in recorded decisions](https://www.arcroomai.com/usecases/meetings-that-end-in-recorded-decisions): Team using an agent to process meeting notes

### Learning

- [A student learning to code with Claude Code](https://www.arcroomai.com/usecases/a-student-learning-to-code-with-claude-code): Student or bootcamp learner

### Agencies and freelancers

- [Twelve clients, no mixed-up memory](https://www.arcroomai.com/usecases/twelve-clients-no-mixed-up-memory): Dev agency with many client repos
- [Back to a client after three months](https://www.arcroomai.com/usecases/back-to-a-client-after-three-months): Freelancer with repeat clients
- [Library maintainers and their decisions](https://www.arcroomai.com/usecases/open-source-maintainers-and-their-decisions): Core team of a widely used library
- [A 48-hour hackathon team](https://www.arcroomai.com/usecases/a-48-hour-hackathon-team): Three people, three agents, one weekend
- [Handing the project to the client's team](https://www.arcroomai.com/usecases/handing-the-project-to-the-client-s-team): Agency finishing a build
- [Delhi's finished work, not redone in the US](https://www.arcroomai.com/usecases/an-overnight-relay-between-time-zones): Team split between India and the US
- [A QA agent separate from the build agent](https://www.arcroomai.com/usecases/a-qa-agent-separate-from-the-build-agent): Shop that keeps building and testing apart

### Cost, safety and control

- [Where did the tokens go?](https://www.arcroomai.com/usecases/where-did-the-tokens-go): Founder paying for several agent plans
- [Secrets never leave the laptop](https://www.arcroomai.com/usecases/secrets-never-leave-the-laptop): Any team with .env files
- [Check exactly what was sent](https://www.arcroomai.com/usecases/check-exactly-what-was-sent): Security-minded team lead
- [Switch it off for one sensitive repo](https://www.arcroomai.com/usecases/switch-it-off-for-one-sensitive-repo): Team with one repo under NDA
- [Memory that expires on purpose](https://www.arcroomai.com/usecases/memory-that-expires-on-purpose): Team whose room got noisy
- [ArcRoom down, agents still running](https://www.arcroomai.com/usecases/arcroom-down-agents-still-running): Anyone who can't afford a blocked session
