Up: promotion Available now
Validated insights move up only with approval. To Team when at least 3 people independently confirm it, to Project by a project lead, to Organisation by governance. Personal memory is private and is never promoted.
The AI Hive by IT Labs India: Shared Cognitive Memory Layer
Cognitive memory that stays with you, whichever AI you use.
Change AI vendor, or use several at once. Your company's memory stays yours and keeps working.
One governed memory for your company's people and their AI assistants, organised by organisation, project, client, team and person, permission-checked on every request, and kept up to date for everyone at once.
The problem
You are paying for intelligence and getting amnesia.
You have already invested in AI. Your teams use it daily. Yet most of that intelligence evaporates every time a session ends.
Most models have memory. Almost none have governed flow.
Without a shared, governed layer, every assistant is a stranger to your projects, your clients and your decisions. The people who do remember are the ones who eventually leave.
| What happens now | What it costs |
|---|---|
| Each AI session starts cold | Repeated context-setting and wasted time |
| Knowledge stays with individuals | Staff turnover erases know-how |
| Agents cannot coordinate | Duplicate or conflicting work |
| No record of what the AI knew | Hard to explain decisions in regulated work |
| Personal and team data are mixed | Privacy and leakage risk |
How it works
Memory with structure: scopes, and three governed flows
The AI Hive organises what your company knows into scopes and moves knowledge between them under clear rules. Nothing reaches a shared scope by accident.
FIG. 1 · MEMORY SCOPES AND FLOWSSEVEN SCOPES
Down: inheritance Available now
When a higher scope changes, every affected person's AI gets the update. New members inherit context immediately.
Across: synthesis Available now
Team knowledge can be merged into project knowledge. It is off until an admin switches it on, and always needs approval.
Key capabilities
Everything a shared memory needs to be trusted
These capabilities are built. Most are running on our pilot deployment; those marked In testing are still being validated. The full list, with what is still on the roadmap, is on the features page.
Context pack Available now
What an AI app receives at the start of a task: the memory inherited for its place, with a manifest for cheap refresh. Less repeated context-setting.
Hybrid search Available now
Meaning plus keyword search, with every result re-checked against permissions before it is returned.
Governed writes Available now
Nothing reaches a shared scope without a role, a time-limited grant or an approval.
Client walls Available now
A client's memory is only returned while working on that client or a project linked to it, never to another client.
Audit log Available now
Every allow and deny is recorded with who, where, which app and which model. "What the AI knew" shows the memory versions behind each answer.
Multi-vendor AI access In testing
Change AI vendor, or use several at once: ChatGPT, Gemini, Copilot, Cursor, Codex, local models and more can connect over MCP and REST. Claude Code and Claude Desktop are available now. Built and in testing; anything that does not pass testing will be removed.
Agents working together In testing
AI agents coordinate one task in rooms with an organisation lobby and claims, under the same permissions and audit as memory.
The AI Hive on your phone Available
The web console installs on Android and iPhone from the browser, with opt-in push notifications that never carry memory text.
Read-only connectors Available now
Jira Cloud, PostgreSQL views, IMAP mailboxes and Git repository addresses feed proposed facts into an approval queue. The AI Hive never writes into outside systems.
Built for governance
Deny by default, enforced where the data lives
Row-level securityPostgreSQL row-level security on every table, so tenant isolation is enforced by the database.
Deny-by-defaultA permission service that checks every request. An agent never has more rights than its user.
Audit in the transactionEach audit record is written in the same transaction as the change it describes.
Data, not instructionsMemory is returned labelled as data. Text that looks like instructions is flagged and cannot be bulk-approved.
Who it is for
A fit when many people and many agents share the same work
A strong fit
- Mid-size and large companies where many people and AI agents share projects
- Consultancies and suppliers with many clients that must be kept apart
- Regulated or export-controlled industries, such as engineering, shipbuilding, automotive, finance and health
- Companies using several AI tools
- Long programmes with staff turnover
Probably not a fit
- Very small teams
- Companies that barely use AI
- One-off pieces of work
If that is you, a simpler tool will serve you better, and we would rather say so now.
What you need to run it
Run by us in your cloud, or on your own server
The AI Hive runs on our pilot deployment today. For your pilot, where it runs is agreed in the scoping workshop. Your data stays in your infrastructure, and fact extraction uses a small model on your own server.
Run by us in your cloud
Our recommended first step. We operate The AI Hive inside your own cloud account.
Available now
Self-hosted on your premises
You run it yourselves with Docker Compose today. An air-gapped bundle is on the roadmap.
Docker Compose: available now
Air-gapped bundle: roadmap
Hosted by us
We run it for you. This option is not available yet.
Roadmap
From your side we ask for a sponsor who can decide, one or two teams who use AI regularly, access to the systems you want connected, and a named person to approve promotions. Deployment options and what we ask of design partners
How a pilot works
- 01Scope. A scoping workshop agrees your memory scopes, who approves what, the success metrics, the systems and AI apps to connect, and the deployment option.
- 02Deploy and connect. We install The AI Hive, create your organisation structure, people, roles and grants, and connect the agreed systems and AI apps.
- 03Run with a first group. A first team uses shared memory in daily work. Approvals, promotion and the audit log are exercised for real.
- 04Review. We review the agreed metrics together and decide on next steps. Stopping is a valid outcome.
The plan is agreed per pilot. Cost is scoped per pilot, so talk to us. Pilot programme
Ready today / next
What is built, and what is not
The full lists are on the roadmap page. Nothing here is a promise of a date.
Ready today
Available now
- Seven memory scopes with promotion, inheritance and synthesis
- Context pack, hybrid search and versioning
- Governed writes, client walls and the audit log
- Claude Code and Claude Desktop
- Read-only connectors: Jira Cloud, PostgreSQL views, IMAP mailboxes, Git repository addresses
In testing
- Multi-vendor AI access over MCP and REST
- Spaces registry and space mapping
- Agents working together in rooms
- Microsoft 365, Google Drive and GitHub connectors
Built and in testing. Anything that does not pass testing will be removed.
Next
Roadmap
- SSO with Microsoft and Google, and SCIM
- Customer-managed keys (BYOK)
- PII detection and masking
- Hash-chained, exportable audit
- Retention policies per scope
- Helm chart and air-gapped bundle
- Hosted option
- Independent penetration test and SOC 2 readiness work
Live pilot
Honest status
The AI Hive runs on our pilot deployment. Features marked In testing are built and being validated; anything that does not pass testing will be removed. We're onboarding a small number of design partners. This is early access, not a self-serve product, and the roadmap is published openly so you can see what is built and what is not.
Memory is not a feature. It is the foundation.
See how The AI Hive could work in your organisation
We are onboarding a small number of design partners to the pilot programme. Tell us about your teams and the AI tools they use, and we will scope a pilot together.