Hivemind
By Lesur.AI
Open-source collective memory for your AI agents: share the context of your projects across assistants, tools and sessions, and retrieve your knowledge along with its sources, via the MCP protocol.
Overview
Hivemind provides AI agents with a shared memory so they can pick up where they left off without having to start from scratch. Decisions, observations and documents remain accessible from one session to the next, even when you switch assistants or model providers.
Three complementary levels of memory
- SHORT — Working notes: time-stamped comments, decisions, queries and tasks, assigned to each staff member.
- MID — Project report: Consolidating notes into structured Markdown documents to make it easier to pick up where you left off.
- LONG — Related topics: semantic search across two distinct and complementary sets:
- Memory Vault: historical records from the project’s memory, preserved during consolidation operations and subsequently indexed. To retrieve a past decision, a lesson learnt or a context that is missing from the current summary.
- Documentary record: documents that you import yourself — procedures, reports, publications or technical documentation. To search for relevant passages and link the knowledge they contain.
Verifiable knowledge
Search results are linked to their source passages. For the Memory Vault, the provenance also identifies the historical snapshot, the original file and the date of capture; a content fingerprint enables the integrity of the data to be verified.
This enables your agents to back up their responses with verifiable references and to distinguish between historical knowledge and the current context. This traceability proves the origin of the content, although it does not in itself guarantee its accuracy or current validity.
Practical applications
- Communicate the background and decisions relating to a project amongst staff and teams.
- Capitalise on resolved incidents and lessons learnt from operations.
- Identify the reasons behind a decision and the sources that support it.
- Compare the project experience with your reference documents.
In your environment
Hivemind is open source under the Apache 2.0 licence and can be self-hosted; it offers dedicated access for agents and a web portal for viewing the memory and monitoring processing. The location of AI processing depends on the conversation model and embedding services you configure.
Find out more: Hivemind presentation · GitHub project
How to use
See the Quick Start to set up the environment, start the services using Docker Compose, create your first workspace and connect your MCP agents.
The prerequisites include S3-compatible storage and services for conversation models and embeddings. For a long-term installation, follow the deployment guide corresponding to the version installed.
Support
To explore a use case, please use the Marketplace contact form or contact bonjour@lesur.ai.
Community support is available via the GitHub issues, according to the support policy, with no guarantee of a response time.