Murmurent¶
▶ Watch the 7-minute getting-started video (install, the agents, and joining a lab or a centre).
Murmurent is an agentic AI operating system that provides:
- Reference agents: a set of specialist agents (literature search, computation, adversarial review, security audit, and others), designed and initialized for biomedical and basic life-science research, that you delegate to by name.
- An Obsidian-based vault: your private, git-backed knowledge base that survives across sessions and projects. It holds your Oracle (personal and lab-shared structured memory), your daily lab notebook, and your own notes and maps.
- Hard rules and hooks: data-governance guardrails enforced in software. Directories can be marked immutable (to protect sensitive raw data) or append-only (to accumulate analyses without overwriting them), and there are guardrails against committing sensitive data such as passwords and clinical records.
- Cryptographic membership: labs, cores, and projects are held together by signed identity certificates (centre root → PI → lead → member).
- The commons: independent groups sharing a common set of agents, rules, and infrastructure, so institutional knowledge accumulates across projects and personnel instead of being trapped on one machine.
Murmurent is built on Claude Code today, although nothing in its design is tied to it: the same tiers, agents, and governance rules could run on another agentic AI system.
You can use Murmurent on your own. You do not need a lab or a centre to participate: the reference agents, the vault, and the governance rules all work for a single researcher on one machine. A lab or centre adds shared memory and infrastructure on top, when you want it.
Where to start¶
| You are… | Start here |
|---|---|
| New: what does this add over plain Claude Code? | Getting started |
| Installing on your machine | Install & setup |
| Using Murmurent in a directory you already have | Making a repo Murmurent-ready |
| Starting a project (in your lab, or across labs) | Creating a project |
| A PI setting up your lab | Group Slack setup · The lab-mgmt repo |
| A mayor bootstrapping a centre | Centre Slack setup · Public directory setup |
| Wondering how the trust model works | Membership IDs & the trust chain |
| Looking for a command | CLI manual |
| Looking for a published choreography to install | The murmurent_public directory keeps the index of all public choreographies in choreographies.tsv; see Choreographies |
How a centre is organized¶
Murmurent is structured around four types of entities:
- Individual members: each person has their own membership identity, their own agents, and their own personal vault.
- Groups: a group is either a lab (a research group) or a core (a shared facility such as a proteomics or imaging centre). Every group is led by a PI and owns its own members, data, and workflows.
- Projects: a project is a unit of work that brings individual members together around shared repositories and data. Its members can come from a single group or from several groups at once.
- The administration: the centre-level layer that maintains the registry of groups and projects and issues the identity certificates that bind members, groups, and projects together.
Each group documents and runs its own workflows on top of the shared agents and rules, keeping authority over its own people and data, while the administration maintains the shared registry and trust chain.
For the full list of the repositories and local files Murmurent uses, see What Murmurent creates.