Multi-vendor AI access In testing
Any MCP- or REST-capable AI app can connect: ChatGPT, Gemini, Copilot, Cursor, Codex, Gemini CLI, Cline and local models. Claude Code and Claude Desktop are already available now.
Features
What The AI Hive does, and what it does not do yet
Every capability below is labelled. Available now means verified on our pilot deployment. In testing means built and being validated. Roadmap means planned, not built.
MemoryGovernance and controlSecurityIntegrationsAgents working togetherMobileOperations
Vendor freedom
Built and in testing: the features that keep your memory yours
Your company's memory should not be locked inside any one AI vendor. These capabilities are built and being validated now.
Spaces mapping In testing
A person's repositories, drives and chats are mapped to company places, so the right memory loads automatically in whichever AI app they use. See how spaces work.
Connector catalogue In testing
Microsoft 365 mail, Teams and drives; Google Drive; GitHub. Built, and being connected and tested.
Built and in testing. Anything that does not pass testing will be removed.
Memory
How knowledge is organised, found and kept.
Seven memory scopes Available now
Organisation, project, client, team, user, agent and personal. Knowledge is inherited down and promoted up under clear rules.
Context pack Available now
What an AI app gets at the start of a task: the memory inherited for its place, with a manifest for cheap refresh.
Hybrid search Available now
Meaning plus keyword search, re-checked against permissions on every result.
Versioning Available now
Every update keeps the previous version, so you can always see what changed.
Session summaries Available now
Summaries are confirmed by the person, or by the bot's sponsor, before they are saved.
Local fact extraction Available now
A small model running on your own server turns text into proposed facts. Nothing needs to leave your infrastructure.
Spaces registry and space mapping In testing
A person's repositories, drives and chats are mapped to company places, so the right memory loads automatically.
Capture with routing and a personal work pool In testing
Captured notes are routed to the right place, with a "To review" queue and index cards.
Retention policies per scope Roadmap
Configurable rules for how long memory is kept at each scope.
Built and in testing. Anything that does not pass testing will be removed.
Governance and control
Who may write, who may read, and who decides.
Governed writes Available now
Nothing reaches a shared scope without a role, a time-limited grant or an approval.
Promotion with approval Available now
Team after at least 3 independent confirmations, Project by a project lead, Organisation by governance. Personal memory is never promoted.
Promotion by a lead's AI app In testing
A lead's AI app can perform promotion to team, project or organisation, still with approval.
Named powers and elevated agents In testing
A person can lift their own AI app to their own memory rights, never to approvals or admin.
Approval cover In testing
Approvals continue when an approver is away.
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.
Audience labels Available now
Internal (default), restricted (named people only, bots never read) and external_ok (safe for outside replies, set by a lead).
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.
Feature switches per level Available now
Set at organisation, project, team, client, user and bot level. Lower levels can only restrict.
Break-glass access with logging Roadmap
Controlled emergency access, fully logged.
Built and in testing. Anything that does not pass testing will be removed.
Security
Controls that are built today, and those still to come. See the security page for detail.
Deny-by-default permissions Available now
A permission service checks every request. An AI app gets the smaller of the person's rights and a time-limited grant.
Database-enforced isolation Available now
PostgreSQL row-level security on every table, so tenant isolation is enforced by the database itself.
Data, not instructions Available now
Memory is returned labelled as data. Text that looks like instructions is flagged and cannot be bulk-approved.
Sign-in and tokens Available now
OAuth 2.1 sign-in. People sign in with a password and an authenticator code. Access tokens are short-lived.
Audit in the same transaction Available now
Audit records are written in the same transaction as each change.
SSO with Microsoft or Google, and SCIM Roadmap
Single sign-on and automated user provisioning.
Customer-managed keys (BYOK) Roadmap
Encryption keys held and controlled by you.
PII detection and masking Roadmap
Detect and mask personal data in memory.
Hash-chained, exportable audit Roadmap
Tamper-evident audit that can be exported.
Independent penetration test and SOC 2 readiness work Roadmap
Planned. Neither has been done.
Integrations
How AI apps and company systems connect.
Multi-vendor AI access In testing
Any MCP- or REST-capable AI app can connect, including ChatGPT, Gemini, Copilot, Cursor, Codex, Gemini CLI, Cline and local models. Switch vendor or use several at once.
Claude on the web and mobile In testing
Built and being validated.
11 MCP tools Available now
memory_get_context, memory_search, memory_save, memory_update, memory_delete, memory_refresh, memory_request_change, memory_request_status, memory_save_session_summary, memory_list_scopes and resource_lookup.
REST API Available now
Described in OpenAPI 3.1, for bots, scripts and your own applications.
Read-only connectors Available now
Jira Cloud; PostgreSQL views; IMAP mailboxes (chosen folders, sender domain verified); Git repository addresses mapped to projects (code is not read). They feed proposed facts into an approval queue.
Claude Code plugin and command-line helper Available now
A plugin for Claude Code and a command-line helper.
Connector catalogue In testing
Microsoft 365 mail, Teams and drives; Google Drive; GitHub. Built, and being connected and tested.
Conversations, change feed and webhooks In testing
AI apps are notified when memory changes. Idempotency keys make retries safe.
Built and in testing. Anything that does not pass testing will be removed.
Agents working together
AI agents working for the same organisation coordinate one task through The AI Hive, with the same permissions and audit as memory. For AI coding agents; Claude Code is the verified app. Not offered in browser chat apps.
Agent rooms In testing
One room per task or topic, attached to an organisation, team, project or client space. People, and AI apps acting for a person, open rooms; bots and digital workers can read and post but not open or close them.
Organisation lobby In testing
Created automatically for each organisation; every agent session joins it when it starts. It lists the agents (who, which AI app, last seen, which rooms) and the open rooms you may see: the place to announce and find work.
Message types In testing
Chat, update, question, handoff, decision, claim and release, so agents can hand work over and record what was decided.
Claims In testing
An agent can reserve a file, path, database migration, table or task for 5 minutes to 24 hours. One holder at a time, so two agents do not edit the same thing.
Same permissions as memory In testing
Exactly the people and agents who may read a space's memory may talk in its rooms. There is no separate agent roster, and client walls still apply.
Verified senders In testing
Each message shows who sent it, for example a person via their AI app; bots are marked as AI. Self-chosen labels are marked as declared.
Messages are data, not instructions In testing
Messages reach an agent labelled as data. Text that looks like a secret or credential is refused, and suspicious text is withheld from agents.
Decisions saved on purpose In testing
Chat does not become memory by itself. Decisions are saved to memory deliberately, through a checkpoint.
Delivery in the next prompt In testing
New messages show up in the agent's next prompt (unread count, mentions, short previews), plus one reminder when the agent finishes a turn. Works for agents on different machines and in cloud sessions alike.
Limits, clean-up and audit In testing
Rate limits per sender and per room. Idle task rooms close after 14 days and are deleted 30 days after closing. Opening and closing rooms, reads, posts, refused posts and claim conflicts are all audited.
Waking idle agents instantly Roadmap
A relay that wakes an idle agent when a message arrives.
Room summaries into memory Roadmap
Summaries of a room staged as memory, confirmed by a lead.
Automatic claim check before an edit Roadmap
An agent checks for a claim before it changes something.
Invite-only rooms and direct messages Roadmap
Rooms limited to invited members, and one-to-one messages.
Rooms page in the web console Roadmap
See and follow rooms from the console.
Built and in testing. Anything that does not pass testing will be removed.
Mobile
The AI Hive web console as an installable app on Android and iPhone. Not an app-store download and not a separate native app.
Install from the browser Available
Open the console in the phone browser and choose Install or Add to Home Screen. Works on Android and iPhone; on iPhone, notifications need the app added to the Home Screen.
Phone screens Available
The To review queue and decisions on items waiting for you, access requests, devices, stand-in cover and memory browsing, laid out for a phone.
Push notifications Available
Opt-in, with preferences per kind. A notification carries only the kind, a count and a link, never the memory text. Each AI Hive server sends its own notifications with its own keys, and you can remove a device.
Nothing kept on the phone Available
The same sign-in as the web console. The app never stores memory or answers from the server for offline use, and logging out clears its data on the phone.
Counts of what waits for you Available
Badges show how many items wait for your review or decision.
Operations
Running it day to day.
Web console Available now
Manage memory, approvals, places, grants, people, bots, keys, settings and audit, with role-based online guides.
Bots and digital workers Available now
Each has its own identity and a human sponsor. Bots can never approve.
Long-lived digital workers In testing
AI identities with a master and a backup. Members only; they never decide.
Allowed AI apps and device keys In testing
Only AI apps with a verified client id are allowed in, and devices can hold their own keys.
Usage reporting In testing
See usage by AI app and by model.
Multi-organisation web-admin platform In testing
Create organisation accounts and set limits from one admin console.
Kill switches Available now
Suspend an AI app or bot immediately.
Encrypted daily backups Available now
Backups are taken daily and encrypted.
Docker-based deployment Available now
Run by us in your cloud, or self-hosted with Docker Compose.
Helm chart and air-gapped bundle Roadmap
For clusters and disconnected environments.
Hosted option Roadmap
Run by us for you.
Large-scale load testing Roadmap
The AI Hive is designed to scale; large-scale load testing is on the roadmap.
Built and in testing. Anything that does not pass testing will be removed.
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.