skills
- python-convention/SKILL.md
- frontend-patterns/SKILL.md
- commit-convention/SKILL.md
- gitflow/SKILL.md
data sources
- Backend DB read-only
- Backoffice DB read-only
- CRM no access
embeddings
- architecture-decisions/ vector index
Agent context control
Centralize the knowledge your AI uses across your organization.
Every person’s agent receives the skills, data sources and document indexes their role allows — versioned once, enforced on the server, traceable to a source. No more prompts pasted into chat or database access handed out one agent at a time.
For Platform, DevEx and Security teams rolling out Codex and Claude Code.
hivemnd delivers a different governed context bundle to each role: the approved skills for that job plus only the data sources and document indexes that person’s permissions allow.
Runs where your security team needs it
SaaS Private instance Self-hosted
The improvement loop
Centralize the improvement loop.
The best knowledge shows up while the work is happening. In hivemnd anyone can propose it against the exact version they used, a maintainer reviews it line by line, and it ships as a new release every agent picks up on its next sync. Every lesson compounds instead of staying in one person’s head.
--- name: sales-playbook version: 2.4.0 --- # Sales playbook ## Discounts Approve up to 15% with sign-off from the sales lead. ## Handoff to onboarding Book the kickoff within 48 hours. - 01
Anyone can contribute
A teammate or an authorized agent proposes an improvement against the immutable version they actually used.
- 02
Maintainers review the contribution
A line-numbered diff, the justification and an append-only review history. Nobody approves their own proposal.
- 03
One approval ships it to everyone
Approval publishes the candidate and the next cumulative release atomically. The CLI syncs it to every workspace.
The missing layer
Your company knows how the work is done. Your agents don’t.
Prompts live in chat threads. Skills live in personal dotfiles. Database access gets granted one agent at a time. Every team ends up running a different, unverifiable version of your company.
The same question, three answers
Three people wrote three versions of the rules, so three agents work three different ways — and nobody can tell which one is current.
Access grows, nobody can explain it
Every new client and connector becomes another place to reason about identity, permission and revocation.
Oversight starts by reading your team
Most visibility tools begin with transcripts. hivemnd begins with system evidence and stores no prompts by default.
How it works
It is not magic. It is specialized infrastructure for distributed agents.
You publish company knowledge once. Every machine that joins pulls the version it is authorized to have, on a schedule, into the agent client its owner already uses.
01 / PUBLISH
Publish approved skills
Version trusted skills and always-on documents in one catalog and release them from a single source of truth.
02 / CONNECT
Connect governed data sources
Expose approved PostgreSQL queries through server-enforced permissions and each person’s effective access.
03 / SUPERVISE
See evidence, not transcripts
Track sync health, access decisions and explicit artifact use without storing prompts or retrieved content by default.
- Preview every local change before it is applied
- Verify hashes and install releases atomically
- Roll back safely when a release is unhealthy
Today it runs natively inside Codex and Claude Code, installing only into the paths it owns and never touching personal configuration.
Using a different one? Tell us which in your access request. If your team needs it, we will look at it and work on the integration with you.
npm install -g @hivemnd-ai/cli
hivemnd init
- Connected to shared.hivemnd.cloud/acme
- 4 skills synchronized to Codex and Claude Code
- MCP server registered for controlled source access
- Automatic sync scheduled every 15 minutes
Isolation and control
You choose how far your knowledge travels from your own infrastructure.
Three isolation levels, one product. Start on the shared platform and move deeper whenever your security team asks for it.
-
SaaS
To get started and try it. Your organization lives in its own isolated workspace inside our shared platform.
-
Private instance
Managed by us, dedicated only to you. Your own deployment, with no neighbours sharing it.
-
Self-hosted
Your data lives on your servers. Not even we have access to it.
Auditable actions
Every access decision and every explicit artifact use is recorded. Prompts and retrieved content are never stored by default.
Profiled knowledge and access
Each person reaches only what their role needs, resolved on the server rather than filtered in the client.
Early access
We are opening hivemnd to strategic users.
We are offering free early access to the platform. Tell us about your use case and we will get in touch to set you up.
- Healthy agent installations
- Answers backed by cited sources
- Blocked unauthorized access
- Stale guidance found before it spreads
Request access
Free while the early access program runs.
Published once
versioned in one catalog and synchronized to Codex and Claude Code
Permission-aware
retrieval governed by each person’s effective access, enforced on the server
Audit without spying
operational evidence without storing prompts or retrieved content by default