Give your AI agents a shared understanding of your work. Hivemind connects your documents, decisions and what agents learn, so the next task starts with the knowledge your group has already built.
Open source · Self-hostable · Works with MCP-compatible tools
“Keep the pilot small. Review accessibility before launch.”
“Accessibility review is complete. Two fixes remain.”
Retain what was decided, why, and what comes next.
Give agents access to findings from the wider group.
Keep your memory when you change assistants or models.
A team, a business, an NGO, a research group or a community: when people and agents share a project, they need access to the same history. Here are four ways to put that memory to work.
Keep requirements, tradeoffs, open questions and progress together. An agent picking up implementation or review can consult the decisions behind the work.
“Where did we leave off, and why did we choose this approach?”
Useful memory: decisions, constraints, completed work and next steps.
Build shared context from incident reports, procedures and agents’ findings. Give the next investigation access to previous resolutions and the conditions in which they applied.
“Have we seen this before? What resolved it?”
Useful memory: symptoms, affected systems, resolutions and exceptions.
Bring papers, experiment notes and working hypotheses into a common memory. Agents can explore related findings and retrieve source context as the research develops.
“Which findings support this hypothesis?”
Useful memory: hypotheses, experiments, evidence and unresolved questions.
Preserve the plans, commitments and lessons of an NGO or community project. Give agents helping new contributors the context accumulated by those who came before.
“What have we committed to, and what still needs attention?”
Useful memory: programs, partners, commitments and handover notes.
Hivemind builds a model of the knowledge it ingests, deriving the kinds of entities and relationships from your source material. This is its automatic ontology: experiments and hypotheses for a research group; programs, partners and commitments for an NGO.
Your agents can retrieve connections alongside source passages. A decision stays connected to the project, people and constraints that give it meaning, without your team having to design that knowledge model by hand.
Read about automatic ontology and our latest evaluation ↗An agent can follow the relationship between a program, its funding and the document describing its progress.
Keep the assistants and coding tools you already use. Connect compatible agents to a shared Hivemind space, with rules that define what the group needs to remember.
Add source documents and set the project’s memory rules. Give agents access to the spaces relevant to their work.
Agents leave observations, decisions and open questions. Hivemind consolidates those notes into a structured project memory.
At the next session, an agent reads the project context and queries connected knowledge, then contributes what it learns.
Memory belongs to your shared workspace, across people, sessions and model providers. Host Hivemind in your own environment, control access to each space and keep the project’s consolidated memory in readable Markdown. Your choice of AI tools can evolve alongside your work.
Agents connect through MCP with scoped access tokens. Use MCP-compatible clients and configure agents to read shared context and record their work. Consolidation and document extraction use your configured model providers; chat and embedding profiles are configured separately.
Provider configuration ↗Teams can share spaces on the same Hivemind deployment. Project Mesh also lets an administrator pair two Hivemind instances around one logical space, starting with one source and one empty target. Each team can retain control of its own infrastructure.
The current pairing supports two instances. Shared writes require acknowledgment from all members; populated spaces cannot be merged.
Project Mesh guide ↗Hivemind is open source under Apache-2.0. The self-hosted stack uses Docker Compose, S3-compatible storage and configured chat and embedding providers. You operate the deployment and pay for the infrastructure and inference you use.
Deployment requirements ↗Choose a project where decisions and handovers matter. Connect your agents and give the next session something useful to build on.