Getting started
Does this work with AI tools other than Claude?
Yes. Second Brain is an MCP server, so any MCP-compatible client can connect to it, including Claude, Claude Code, Cursor, Codex, ChatGPT and others. Exact menu names and plan availability change over time, so check the connect guide for the current methods. The Obsidian plugin also syncs your vault notes, making them available to every connected client.
Do my devices sync? How does multi-device work?
There is nothing to sync. You run one Worker in your own Cloudflare account, and every install, meaning the desktop app, the CLI, the browser tools, the Obsidian plugin and each AI client, is a door into that same brain. Save something on your laptop and it is already there when you recall it on your phone. Connecting a new device just means pointing it at the same address and signing in. Setup starts in the setup guide.
Is this production-ready?
It is stable for personal use and small teams. The desktop app handles setup without technical steps, and Team Edition isolates every member's personal workspace, uses individual credentials, author-locks shared memories, and records administrative and sharing activity. It is still self-hosted software: the owner is responsible for the Cloudflare account, deployment, backups, team membership and any compliance requirements. See the teams guide for how sharing works.
Why not just use my AI tool's built-in memory?
Built-in memory, such as Claude's, is a black box. You cannot search it, audit it, or control what gets stored, and it only works inside that one product. Second Brain shows you every memory in a dashboard, lets you append, edit or delete anything you did not intend to keep, and works across Claude, ChatGPT, Cursor and any other MCP-compatible client.
How do I contribute or suggest features?
Open a GitHub Discussion rather than an Issue. Discussions are for feature requests and ideas, while Issues are for bugs with steps to reproduce. Discussions is also where releases get announced and where the roadmap gets shaped.
Your data and privacy
What happens to my data?
Everything lives in your own Cloudflare account: the D1 database, the Vectorize index, KV and the Worker. Cloudflare provides the infrastructure; the Second Brain project and its maintainers have no access to your deployment, your memories or your credentials. Integrations contact only the external services you explicitly connect.
What is my bearer token (auth token)?
For the owner, it is the password chosen during setup, stored as AUTH_TOKEN. The desktop app calls it a password and the API documentation calls it a bearer token, but they are the same owner credential. On a team brain, every member receives a separate one-time sign-in token instead, stored only as a hash, so an admin can reset it but can never reveal it again. Never place either kind of token in a URL. The connect guide covers where it goes for each client.
Costs
Does this eat a lot of tokens?
No. A recall call at the start of a conversation adds roughly 500 to 2,000 tokens depending on how many memories come back, which is negligible against a 200k context window, and you can tune topK down to minimize it. Store calls are tiny, effectively free. The one habit worth avoiding is letting an assistant store its own response after every turn; store the responses that contain something reusable, such as a decision or a solution, not every reply. The costs guide has the full breakdown.
Does Team Edition cost extra?
No. There is no separate Second Brain team service or deployment; the same Worker supports solo and team use. Your practical capacity and cost depend on the D1, Vectorize, Workers AI, KV and Worker usage in your own Cloudflare plan. A small team can often stay within free-tier budgets; monitor your Cloudflare usage as the number of people and memories grows.
Teams
Can an admin read another person's private memories?
No. An admin can manage membership, credentials, defaults, integrations and the activity log, and none of those privileges grant access to another person's personal workspace. Admins can see a count of another person's private memories, which exists for offboarding decisions, but never the content. Nobody on the team, including the owner, can read anyone else's personal layer. The teams guide compares personal and shared layers directly.
Who can edit a shared memory?
Everyone on the team can read it. Its author and team admins can edit, append, delete, change its status, or move it back to Personal. Other members cannot quietly rewrite someone else's shared decision.
Where do new memories go on a team brain?
An explicit workspace choice on the capture wins. Otherwise Second Brain uses the member's own override, if one is set, then the organization default. The dashboard calls that inherited choice Auto, and it shows where Auto currently resolves before the capture is sent, so the destination is never a guess.
What happens when a member is removed?
Their access is revoked and their personal memories are permanently deleted. Memories they authored in the shared layer remain, with their name still attached, because those memories already belong to the team. Suspension is the reversible alternative: it blocks access without deleting anything.
Troubleshooting
What about stale or outdated memories?
Use append when something changes: it adds a timestamped update to the existing memory, so the history is preserved and the most recent state ranks first. Use update when the old content should be replaced entirely, and forget to remove something completely. You rarely need to manage this by hand, because recall already deprioritizes older memories, and storing an updated version often resolves the contradiction for you automatically. See your memory, day to day for all three in use.
Technical details
Why not just use local files or my Obsidian vault?
Local files work well when all your AI tools share one machine and you have a small number of notes. As your memory grows, you end up stuffing whole files into the context window hoping the right one is in there, and files do not travel with you to Claude on your phone or another machine. Second Brain retrieves only what is relevant to the current conversation, searches by meaning, and is available from anywhere because it lives in your own Cloudflare account.
Why embeddings instead of a virtual filesystem?
A filesystem approach stores memories as files the AI reads directly, which works at small scale. But search becomes keyword or file-path based, and context costs rise with every memory you add. Embeddings convert every note into numbers representing its meaning, so a search for "onboarding problems" finds "users dropping off at checkout" with no keyword overlap at all. Second Brain handles the embedding machinery automatically; you do not need to think about it.
Can I use a different embedding model?
Yes. Community members have already swapped in Qwen with 1024 dimensions for better multilingual performance. To change it, update the embedding model name in the Worker and change the Vectorize index dimensions to match. The shipped default is bge-small-en-v1.5 at 384 dimensions.
How does recall know what is relevant?
Every memory is embedded as a 384-dimension vector using bge-small-en-v1.5 on Cloudflare Workers AI. When you recall, your query is embedded the same way, and Vectorize finds the closest stored vectors by cosine similarity. Recall works by meaning rather than keywords, so you never need to remember how you phrased something when you saved it. The your memory guide covers how ranking and reconciliation build on this.