Context repetition
You lose time re-explaining information another AI already received.
Universal memory across AIs
All your AIs.
All your AIs.
One single memory.
Connect ChatGPT, Claude, Codex, Claude Code, Gemini CLI, Grok, Antigravity and other MCP-compatible clients to the same external brain. What one session records can be searched and recalled by the others — across models, projects and tools.
One brain that follows you across different AIs
AI Brain is built on MCP, an open protocol that standardizes how AI applications connect to tools and external data sources. Instead of keeping a separate memory inside each platform, you connect different clients to the same endpoint.
Availability of tools, OAuth authentication and write actions may vary depending on the client, the plan and the permissions each platform offers.
The problem
Your AIs are powerful.
Your AIs are powerful.
The memory between them is still fragmented.
You discuss a strategy in ChatGPT. Then you open Claude to go deeper on the reasoning. You use Codex to implement. You ask Gemini about another part of the project. You ask Grok to research a reference. Every tool produces decisions, findings and learnings that matter — but when you switch clients, most of the context has to be rebuilt.
You explain all over again:
- Who you are
- What you are building
- Which decisions were already made
- Which approaches were already discarded
- Which patterns must be followed
- Which mistakes must not be repeated
- How a given project works
- What your communication tone is
- Which rules are permanent
- What another AI already figured out
The problem is not a lack of intelligence in the models.
The problem is the lack of a common memory between them.
You don't need one more isolated AI. You need the AIs you already use to share continuity.
Scattered decisions
An important decision may be buried in an old conversation, an issue, a note, a commit or a document.
Platform dependency
Context stays trapped inside the ecosystem where it was created.
Disposable learning
One AI solves a problem, but that learning never becomes reusable knowledge for the others.
Fragmentation across projects
Patterns that should apply to several projects end up duplicated or forgotten across different repos and conversations.
AI Brain turns temporary context into reusable memory.
The solution
An external brain,
An external brain,
reachable by all your AIs.
AI Brain is a hosted MCP server that works as an external memory layer. Every user has their own individual brain. Once you connect an AI to AI Brain, it gains access to tools built specifically for memory.
- Store memories
- Search knowledge
- Recall a complete note
- Explore recent memories
- Remove information
- Navigate links and backlinks
- Reuse decisions across projects
- Share context between different clients
The memory does not belong to ChatGPT. Not to Claude. Not to Codex. Not to the current project. It belongs to the user.
The AI can change. Your external brain stays.
Each client makes its own connection to the MCP server. Once information is stored in AI Brain, it stops existing only inside the conversation that produced it. Another connected AI can search that knowledge, recall the full memory and use it as context for a new task.
One session discovers. AI Brain preserves. Another session continues.
// a concrete example 1. Claude analyses a product's architecture. 2. Claude records the decision in AI Brain. 3. Days later, Codex starts implementing. 4. Codex searches AI Brain. 5. Codex recalls the architectural decision, the constraints and the reasoning behind it. 6. Implementation starts with the context already consolidated.
How it works
Six steps between subscribing and compounding memory.
Create your brain
Subscribe to AI Brain and sign in with the subscription email to activate your individual brain, tied to your account.
Connect your AIs
Add the AI Brain endpoint to your MCP-compatible clients.
https://mcp-ai-brain.vercel.app/mcp
Authenticate
On clients with native OAuth support, sign-in happens through AI Brain's own consent screen: you provide the email tied to the account, receive an OTP code and authorize the connection.
Record knowledge
Ask the AI to store a decision, a pattern, a rule, a project summary, a learning, a preference, a procedure, an architecture, a piece of documentation or any important context.
Recall it from any connected client
In another AI, ask it to search for or recall that memory.
Build a compounding memory
Over time, AI Brain stops being just a set of notes and becomes a graph of related decisions, projects, people, concepts and learnings.
The five MCP tools
Five memory primitives.
Five memory primitives.
An entire layer of continuity.
AI Brain does not expose dozens of generic tools. It offers five clear operations, designed specifically for memory.
| Tool | Function |
|---|---|
brain_store | Creates or updates a memory by title |
brain_recall | Recalls a complete memory |
brain_search | Searches the brain and returns ranked results |
brain_list | Lists recent memories, optionally filtered by tag |
brain_delete | Removes a memory by title or identifier |
Store
Writes or updates a Markdown note: up to 64 KiB of content, up to 20 tags, [[wiki-link]] links, decisions, rules, context and documentation. Every time a memory is written, its wiki-links are parsed and turned into connections in the graph.
Recall
Brings back a complete memory by title or identifier. The response can include full content, metadata, tags, outgoing links, backlinks and relationships to other memories.
Search
Searches title and content using Postgres full-text search, with ranked results, relevant snippets and tag filters. Current search is lexical — semantic search is planned evolution, not a feature of the current plan.
Explore
Lists recent memories with optional tag filtering. Useful to explore the brain, review recent records, find memories by category and understand how a project evolved.
Forget
Removes a memory by title or identifier. The operation is idempotent and derived relationships are updated in cascade.
Intent, not a file path
Five cognitive verbs — remember, search, recall, explore, forget — give the model a smaller, semantically clear surface instead of making it reason about folders, paths and file overwrites.
Mental model
Familiar as a second brain.
Familiar as a second brain.
Built for AIs to use.
The mental model draws on connected knowledge systems: Markdown notes, titles, tags,
[[wiki-links]], outgoing links, backlinks, related nodes and a graph.
The difference is that this structure was not designed primarily for a human to open an app
and browse folders — it was designed so models and agents can query memory directly through tools.
It is not just a note. It is a unit of memory another AI can recall and use.
# Decision: async processing for Smart Clips The pipeline must be asynchronous, idempotent, versioned and resumable. ## Reason Videos can run up to 60 minutes and must not depend on the lifecycle of an HTTP request. ## Related - [[Smart Clips]] - [[Job Architecture]] - [[Vercel Runtime Constraints]] - [[Publication Gate]] ## Tags architecture decision smart-clips async-processing
What can be stored
The knowledge that shouldn't disappear
The knowledge that shouldn't disappear
when the conversation ends.
Projects
- Goals and scope
- Architecture and current state
- Next steps and blockers
- Pending decisions
- Relationships to other projects
Software development
- Code patterns and ADRs
- Repository conventions
- Architectural decisions and constraints
- Incident fixes
- Lessons learned and the why behind decisions
Brand and communication
- Positioning and tone of voice
- Bio, taglines, approved phrasing
- Language rules
- Visual preferences
- Audiences, products and offers
Content
- Editorial pillars
- Formats that worked
- Script structures and benchmarks
- Campaign learnings
- Approved answers and recurring objections
Operations
- SOPs, processes and policies
- Checklists and approval rules
- Security procedures
- Client context
- Commercial decisions
Personal knowledge
- Preferences and methods
- Frameworks and studies
- Research and concepts
- People and relationships
- Goals
Benefits
Ten reasons to move memory
Ten reasons to move memory
out of whichever model you're using today.
Continuity across models
Claude to think, Codex to implement, ChatGPT to communicate, another AI to research — the relevant knowledge stays reachable across those steps.
Continuity across sessions
A conversation doesn't have to carry the entire prior history. The AI searches the brain and recalls only the memories relevant to the task at hand.
Continuity across projects
A decision or learning doesn't have to stay locked in the repository where it appeared. Reusable patterns cross projects.
Less context repetition
You reduce the need to manually rebuild the same background in every new conversation.
Less vendor dependency
Your operational memory doesn't have to exist exclusively inside a single model's ecosystem.
Linked knowledge
Tags, wiki-links and backlinks turn isolated memories into a navigable network of knowledge.
Operable memory
The AI doesn't just receive one big file: it searches, recalls, stores and deletes knowledge through specific operations.
Hosted infrastructure
You don't install, host, update or maintain your own memory server.
Governed access
OAuth, per-user isolation and tool policies help control who has access and which operations are allowed.
A foundation for agents and workloops
AI Brain works as the persistent knowledge layer for agents, copilots and Governed Agentic Workloops.
Comparisons
Not better at everything.
Not better at everything.
Better at the job it was built for.
Native memories, Obsidian and GitHub MCP solve real problems and remain excellent at what they do. AI Brain solves the fragmentation between those ecosystems.
Native memory belongs to the client. AI Brain belongs to your architecture.
Native memory can be useful inside a specific platform. The problem shows up when you use several AIs: ChatGPT's memory doesn't automatically become Claude's memory, and Claude's context doesn't become Codex's memory. AI Brain externalizes that layer.
| Dimension | Native memory | AI Brain |
|---|---|---|
| Scope | Usually limited to the product | Cross-client |
| Architectural ownership | Inside the platform | The user's external layer |
| Access by other models | Usually not | Yes, via connected MCP clients |
| Structure | Defined by the vendor | Markdown, tags, links and backlinks |
| Explicit search | Depends on the platform | brain_search |
| Full retrieval | Depends on the platform | brain_recall |
| Explicit write | Depends on the platform | brain_store |
| Use by agents | Limited to available integrations | Available over MCP |
| Cross-project | Depends on the client | Native to the brain model |
Native memory personalizes one AI. AI Brain connects several AIs to the same memory.
AI Brain does not automatically capture every conversation, nor replace each client's local context. It offers an explicit layer that connected AIs use when information is stored or recalled.
Obsidian organizes knowledge for humans. AI Brain serves memory to AIs.
Obsidian is an excellent tool for humans creating, editing and navigating knowledge in Markdown files, and an Obsidian MCP lets an AI interact with that vault. There is no single product called “Obsidian MCP”: there are distinct community implementations — some rely on the Local REST API plugin, others include tags, frontmatter, backlinks and BM25, others run inside Obsidian itself. The core difference is one of layer: Obsidian MCP exposes a vault; AI Brain offers a memory abstraction.
| Dimension | Obsidian MCP | AI Brain |
|---|---|---|
| Primary role | Give access to the vault | Provide external memory to AIs |
| Centre of the experience | Human and notes | AIs and memory operations |
| Storage | Local Markdown files | Hosted, structured backend |
| Organization | Folders, files, frontmatter | Titles, content, tags, links and backlinks |
| Main interface | The Obsidian app | MCP tools |
| Search | Depends on the implementation | Standardized full-text search |
| Cross-client | Depends on local/remote setup | The product's central goal |
| Cross-project | Depends on vault structure | Native to the brain |
| Hosting | Often local | Hosted |
| OAuth | Depends on the implementation | Built into the service |
| Maintenance | The user's responsibility | Operated by AI Brain |
| Needs Obsidian open | In some implementations | No |
| Best for | Writing and organizing notes as a human | Giving AIs memory continuity |
// with an Obsidian MCP, the AI may have // to reason about: Which folder holds the note? What is the file path? Which file should I open? Is this in the frontmatter? Search by name or by content? Edit one section or overwrite the file?
// with AI Brain, intent is direct: I want to store something. brain_store I want to find a memory. brain_search I want the full memory. brain_recall I want recent memories. brain_list I want to remove a memory. brain_delete
Obsidian is where you organize knowledge. AI Brain is how your AIs turn knowledge into continuity.
It would not be accurate to say AI Brain beats Obsidian at everything — Obsidian is better as a human writing and navigation environment. Syncing between the two is an architectural possibility and future evolution, not a current feature.
GitHub MCP connects the AI to GitHub. AI Brain connects all AIs to the same memory.
The GitHub MCP Server is official, maintained by GitHub, and extremely powerful for querying repositories, reading code, managing issues and pull requests, and following Actions and releases. It solves a different problem: GitHub MCP and AI Brain are not competitors — they occupy different layers of the architecture.
| Dimension | GitHub MCP | AI Brain |
|---|---|---|
| Goal | Query and operate GitHub | Share memory across AIs |
| Domain | Code, repos, issues, PRs, Actions | Projects, decisions, patterns and context |
| Primary source | GitHub | The user's individual brain |
| Data model | GitHub objects and APIs | Markdown memories, tags and links |
| Interface type | GitHub operational tools | Memory primitives |
| Cognitive continuity | Indirect | The central goal |
| Cross-project | Within repos and organizations | Repository-independent |
| Cross-domain | Limited to GitHub | Engineering, marketing, ops, content |
| Best for | Operating development | Preserving decisions and learnings |
| Nature | Vertical connector | Horizontal infrastructure |
GitHub preserves what happened. AI Brain preserves what we learned and why we decided.
Commits, diffs, issues and PRs show what happened. But the most important decision — “we chose async processing because videos of up to 60 minutes must not depend on the lifecycle of an HTTP request” — may be scattered across an issue, a PR, comments and documents. In AI Brain it becomes a canonical memory, with a title, a reason, tags and connections.
// a workloop example 1. Codex uses GitHub MCP to analyse a failing workflow. 2. Codex finds the root cause. 3. Codex fixes the code and opens a PR. 4. The fix is validated. 5. Codex records in AI Brain: "In this monorepo, package X must be built before package Y because of the shared contract." 6. Months later, Claude works on a similar project and recalls that learning. 7. GitHub MCP queries the current repo to check whether the pattern still applies.
AI Brain is better as universal memory. GitHub MCP is better as an operational interface to GitHub.
In a MAPE-K-inspired reading: GitHub MCP is sensors and actuators; AI Brain is the Knowledge layer.
Complementary architecture
AI Brain doesn't need to replace your tools.
AI Brain doesn't need to replace your tools.
It connects the learning they produce.
AI Brain does not need to be the source of truth for code — GitHub remains the source of the
software's current state. It also does not replace the human writing environment. AI Brain fills a
different role: preserving and serving the knowledge AIs need to reuse.
Use cases
Who gains most from a memory
Who gains most from a memory
that crosses tools.
Your company changes tools. Your strategy shouldn't vanish between them.
- Preserve strategic decisions
- Share context across AIs
- Keep positioning, products and priorities
- Reduce briefing rework
The implementation changes. The reasoning behind the architecture stays reachable.
- Store architectural decisions
- Record constraints and patterns
- Reuse solutions
- Share context across Claude Code, Codex and other agents
One AI researches. Another writes. Another produces. All can consult the same context.
- Preserve tone of voice
- Record content pillars
- Keep benchmarks
- Store structures that worked
Each client's knowledge shouldn't be scattered across chats, people and projects.
- Client context
- Communication rules
- Decision history and processes
- Delivery standards and lessons learned
Turn accumulated experience into memory your AIs can query.
- Frameworks and methodologies
- Diagnostics
- Project context
- Recommendations and decision patterns
Agents don't just need to execute. They need to learn what must be preserved.
- Long-term memory
- Cross-workloop knowledge
- Execution results
- Rules and domain context
Before and after
Stop rebuilding context.
Stop rebuilding context.
Start compounding intelligence.
Before AI Brain
ChatGPT knows one part. Claude knows another. Codex knows the current repository. One decision sits in an old conversation. Another sits in an issue. A pattern sits in a file. You have to piece it all together again.
After AI Brain
The relevant information is recorded as memories. Memories carry titles, tags and relations. Connected clients search the same brain. Each AI recalls only the context it needs. New learnings can be stored again.
Infrastructure and security
AI memory needs boundaries too.
AI Brain is a hosted service: you don't maintain containers, configure databases, expose local ports or operate your own MCP server. And because a memory can hold sensitive data, outdated content or even malicious instructions, security here isn't just authentication — it exists in layers.
MCP clients
│
▼
OAuth 2.1 / PAT
│
▼
Hosted MCP server
│
▼
Application layer
│
▼
Supabase Postgres
│
├── memories
├── tags
├── links
├── backlinks
└── full-text search
Components
- Hosted MCP — remote server deployed on Vercel infrastructure.
- Database — Supabase Postgres for persistence, search and relationships.
- Authentication — revocable bearer PAT, OAuth 2.1, PKCE S256, Dynamic Client Registration, refresh token rotation, family revoke and consent.
- Dashboard — web area for sign-in, token management, memory browsing and graph access.
- Visual graph — nodes, links, backlinks, ghost nodes, tag exploration and navigation into memory content.
Per-user RLS
Every query runs in the authenticated user's context, with isolation enforced in the database by Row Level Security. Your brain is individual: every access is tied to your identity.
Service role off the user path
The service-role key does not operate directly on user-scoped paths. The tenancy boundary doesn't rely on the application alone — it also exists in the database.
OAuth 2.1 and PKCE
Compatible clients authenticate through a modern consent flow. Connect without spreading static tokens across tools whenever the client supports OAuth.
Zod validation
Inputs are validated at the application boundaries. Invalid data should not pass silently through the infrastructure.
Logs without PII
Observability avoids exposing personal data and sensitive content. Monitoring the infrastructure doesn't require turning your memory into logs.
Memory as untrusted data
Recalled content is wrapped and marked as untrusted data, reducing the risk of a malicious memory being read as a privileged instruction. A recalled memory informs the model — it does not take control of it.
STRIDE-based threat model, versioned alongside the product.
Persistent memory without governance only makes risk persistent.
Compatibility
Built on an open protocol,
Built on an open protocol,
not for a single model.
MCP standardizes how an AI application discovers and uses tools and external sources. That lets the same AI Brain server connect to different clients without building a brand-new integration for every model. Compatibility is communicated at three levels.
Validated by the product
Clients tested end to end by the AI Brain team.
- Claude.ai
- Claude Code
- ChatGPT.com
- Codex CLI
- Codex IDE
Official client support for custom MCP
Clients whose official documentation allows connecting MCP servers, but whose specific flow with AI Brain still needs full validation.
- Gemini CLI
- Grok
- Antigravity
Potential compatibility
Frameworks, agents and clients that implement remote MCP and a compatible authentication method.
Compatible with clients that accept remote MCP servers.
Availability of read, write, OAuth and propagation across surfaces varies by vendor. Connect each
compatible ecosystem to the same brain — some platforms may propagate the connection across their
own surfaces.
Subscription
You're not subscribing to five tools.
You're not subscribing to five tools.
You're subscribing to memory infrastructure.
The five tools are the interface. The real value is the system behind them — an always-available server, persistent storage, per-user isolation, authentication, indexing, search, graph, dashboard, maintenance and ongoing evolution.
AI Brain Individual · monthly
R$49.90/month
Recurring monthly billing, in BRL.
Create my AI BrainConnect your AIs to the same external brain.
AI Brain Individual · annual 2 months free
R$499.00/year
Equivalent to R$ 41.58/month, in BRL.
Saves R$ 99.80 vs. 12× monthly Get the annual planSame brain, same endpoint, same infrastructure.
Included in either plan
- 1 individual brain per user
- Hosted MCP endpoint
- Access to the five tools
- Markdown memories
- Tags
- Wiki-links
- Backlinks
- Full-text search
- Dashboard
- Visual graph
- OAuth 2.1
- Revocable PAT
- Cross-client use
- Cross-project use
- Continuous updates
AI Brain doesn't sell note storage. It offers continuity across the AIs you actually work with.
Current limits
Clear scope today.
Clear scope today.
Structured evolution tomorrow.
what does not exist yet
- Search uses Postgres lexical full-text search. There is no embeddings-based semantic search yet.
- The initial version has one individual brain per user — no brains shared between people or teams.
- The dashboard lets you view and explore content; writing memories happens through the MCP tools.
- AI Brain does not automatically import conversations: an AI must call
brain_storeto persist information. - It is not a substitute for a real-time operational source — for a repository's current state use GitHub MCP; for a CRM, the CRM connector.
Natural evolutions
- Semantic search
- Embeddings
- Hybrid re-ranking
- Team brains
- Shared memories
- Granular permissions
- Obsidian integration
- Import and export
- Retention policies
- Memory versioning
- Contradiction detection
- Knowledge expiry
- Automatic consolidation
- Learning loops
Indicative roadmap. No dates announced.
Frequently asked questions
What people usually ask before subscribing.
What is AI Brain?
A hosted MCP server that works as an external memory shared across different AIs. It lets you store, search, recall, list and delete memories through standardized tools.
What does “Universal Memory Across AIs” mean?
It means memory lives outside any specific model and can be reached by different MCP-compatible clients.
“Universal” does not mean every tool is connected automatically: each client must support MCP and be connected to the server.
Does AI Brain automatically read all my conversations?
No. AI Brain receives information when a connected AI calls a tool such as brain_store. That gives you more control over what becomes persistent memory.
When one AI stores something, do all the others receive it automatically?
The memory becomes available to every authorized AI connected to the same brain. Retrieval depends on the AI calling brain_search, brain_recall or another suitable tool.
Does AI Brain replace ChatGPT's or Claude's memory?
Not necessarily. It can complement native memories, creating an external layer that doesn't belong exclusively to one vendor.
Does AI Brain replace Obsidian?
No. Obsidian is excellent for humans writing, organizing and navigating notes. AI Brain is better at serving memory directly to AIs. The two can be used together.
Does AI Brain replace GitHub MCP? Can I use both?
It doesn't replace it — and yes, they can be used together. GitHub MCP is better for querying and operating GitHub. AI Brain can store the reasoning behind decisions, the outcomes and the learnings worth reusing.
Which AIs are compatible?
AI Brain can be connected to clients that support remote MCP servers and a compatible authentication flow. The published list separates validated clients from potentially compatible ones — see the compatibility section.
Do I need to install a server?
No. AI Brain is hosted. Some clients may require editing a configuration file or creating a connector, but you never operate the AI Brain backend.
How does sign-in work?
On clients with native OAuth, AI Brain presents its own consent screen. You provide the email tied to the account, receive an OTP code and authorize the connection.
Where are my memories stored?
Memories are stored on AI Brain's infrastructure, using Supabase Postgres.
Can someone else access my brain?
The individual version is designed as one brain per user. The application uses authentication and Row Level Security to separate data between users.
Does AI Brain have semantic search?
Not yet. Current search uses lexical full-text search. Embeddings and semantic search are natural evolution frontiers for the product.
Can I edit memories from the dashboard?
The current dashboard is built for reading and exploring. Writing happens through the MCP tools.
Can I delete a memory?
Yes. The brain_delete tool removes a memory by title or identifier.
Can AI Brain hold incorrect information?
Yes. Like any memory system, it can store content that is wrong, outdated or contradictory. That's why AIs should treat memory as recalled context, not absolute truth.
Does AI Brain guarantee an AI will never forget anything again?
No. AI Brain provides persistence and retrieval: the AI still has to store the information and query memory when appropriate. The accurate promise is reducing context loss and repetition.
How much does it cost?
The AI Brain Individual subscription costs R$ 49.90 per month, or R$ 499.00 per year — the equivalent of two months free compared to the monthly plan.
One brain. All your AIs.
Your intelligence shouldn't start over every time you switch models.
Use Claude to think. Codex to build. ChatGPT to communicate. Gemini to analyse. Grok to research. Your agents to execute. And AI Brain to hold the continuity.
Models change. Tools change. Your memory shouldn't start over.
Built on AISAC
The ecosystem
AISAC Global
Vision, thesis, protocol, academy, credentials and research. Who we are and what we stand for.
in progress
The products
AISAC Cloud
Products and infrastructure built for AI-operated companies. This is where AI Brain lives.
you are here
The people
AISAC Community
Founding Cohort 2026 — the first 100 people building AI systems as code.
live