hyperhive/README.md

hyperhive

a swarm of claude-code agents, each in its own nspawn cage, gossiping over unix sockets. config changes flow as git commits, the operator approves them in a browser, every deploy is a tag. cyberpunk-themed dashboard included. 💜⚡

A host-side Rust daemon (hive-c0re) spawns nspawn-isolated agent containers, runs each one's claude turn loop, and brokers messages between them. A privileged manager agent (hm1nd) drives the swarm — proposing new agents, editing their NixOS modules, escalating ambiguous decisions to the operator. Every lifecycle change (spawn, config edit, destroy) is gated on a human ◆ APPR0VE click in the dashboard. Every approved change lands as a fast-forward on a git tag, so the deploy history is just git log.

Why this exists: claude code is great in one window. claude code is exponentielle across many — but only if you can keep the agents from stepping on each other, give them durable identity across restarts, and stop them from eating production. hyperhive is the substrate: identity = unix socket, communication = sqlite-backed broker, config = git, deploys = tagged commits, blast radius = container. The operator stays in the loop without becoming the bottleneck.

host (NixOS, runs hive-c0re.service)
│
├── operator
│   ├── browser → :7000               hive-c0re dashboard (containers, approvals)
│   ├── browser → :8000 / :8100-8999  per-agent web UIs (live SSE, send, login)
│   └── CLI     → /run/hyperhive/host.sock         JSON-line admin protocol
│
├── hive-c0re  (Rust daemon)
│   ├── lifecycle    nixos-container CRUD + per-agent flake generation
│   ├── broker       sqlite messages + tokio broadcast (powers SSE + wake-ups)
│   ├── approvals    sqlite queue, two kinds: ApplyCommit (config) + Spawn
│   ├── auto_update  rebuilds any container whose recorded flake rev is stale
│   ├── dashboard    axum HTTP + async-form actions + SSE message flow
│   └── sockets      /run/hyperhive/{host,manager,agents/<n>}/mcp.sock
│
└── nixos-containers  (each bind-mounts its socket dir → /run/hive,
   │                   credentials dir → /root/.claude,
   │                   durable notes dir → /state;
   │                   manager additionally gets /agents RW,
   │                   /applied RO (deployed-tag mirror),
   │                   /meta RO (swarm-wide deploy flake))
   │
   ├── hm1nd      hive-m1nd serve : claude turn loop +
   │              MCP (send / recv / request_spawn / kill / start /
   │                   restart / update / request_apply_commit /
   │                   ask / answer / remind) + web UI on :8000
   │
   └── h-<name>   hive-ag3nt serve : claude turn loop +
                  MCP (send / recv / ask / answer / remind + agent-declared
                       extras via hyperhive.extraMcpServers) + web UI
                  on a hashed :8100-8999

The turn loop

One message in, one turn out. The harness pops a single inbox message (Recv long-polls server-side, wakes the instant a broker Sent event fires) → builds a wake prompt → spawns claude --print --continue --output-format stream-json --mcp-config … → streams JSON events into the per-agent SSE bus + a sqlite history db → claude drives any further recv/send itself via the embedded MCP server. --continue preserves the prior session so context spans turns without rebuilding from history every wake. If the turn dies mid-stream, the next wake picks up clean — sessions are durable but cheap to discard.

Operator surfaces

Per agent (:8100-8999): terminal-themed live tail with a textarea prompt; slash commands /help /clear /cancel /compact /model <name> /new-session; granular state badge (idle / thinking / compacting / offline) with age timer + last-turn duration chip + model chip; cancel-turn + new-session buttons in the state row; sticky-bottom auto-scroll with "↓ N new" pill; event history backfilled on page load; collapsible inbox + collapsible journald viewer + collapsible agent.nix viewer per agent on the dashboard; deployed-sha chip per container (read from meta's flake.lock).

Dashboard itself (:7000): the swarm-wide view — every container with status chips, every pending approval with inline diff, every open question with operator-answer affordance, every recent message on a unified live stream. A terminal compose box under the message-flow lets the operator drop messages into any agent's inbox: @name picks the recipient with auto-complete from the live container list, sticky across sends, POSTs /op-send which lands in the broker as {from:"operator", to:<name>, body} — same shape any sub-agent sees as a regular inbox message. No special channel, no out-of-band notification system. If the operator says it, the broker carries it.

The config-edit loop

Inverted flow — agents propose, the operator disposes. The manager edits files under /agents/<name>/config/ — agent.nix is a plain NixOS module function { config, pkgs, lib, ... }: { ... }, and arbitrary sibling files in the commit are preserved → commits → submits the sha via request_apply_commit. Hive-c0re immediately fetches that commit from the proposed repo into the applied repo and pins it as proposal/<id> — immutable from the manager's side from then on. Operator clicks ◆ APPR0VE → hive-c0re fast-forwards applied/<n>/main to the proposal, runs nix flake lock --update-input agent-<n> against the host-wide meta flake at /var/lib/hyperhive/meta/, builds via nixos-container update <c> --flake meta#<name>, and either commits the lock + tags deployed/<id> on success or git restores the lock + annotates failed/<id> with the build error + rolls back applied/<n>/main on failure. Denials leave a denied/<id> annotated tag carrying the operator's note.

Meta's git log is the swarm-wide deploy audit trail (one commit per successful deploy). Per-agent applied repos carry the tag-rich state machine for inside-baseball decisions. The manager sees both — proposed repos ship with an applied remote pre-wired, and /meta/ is RO-bound inside the container — so git fetch applied, git show applied/refs/tags/deployed/<id>, git log /meta, cat /meta/flake.lock all just work without constructing paths by hand. See docs/approvals.md for the full state machine + lock-flow walkthrough.

Structured Q&A

Agents don't have to guess. ask(question, options?, multi?, ttl_seconds?, to?) queues a structured question — default recipient is the operator (dashboard renders a free-text / checkbox / radio form), or pass to: "<agent>" to route the question into a peer agent's inbox instead. The answer arrives later as a HelperEvent::QuestionAnswered { id, question, answer, answerer } in the asker's inbox; peer recipients respond via answer(id, answer). Plus remind(message, delay_seconds | at_unix_timestamp, file_path?) for self-scheduled wake-ups when an agent wants to nudge itself later without holding a connection open.

Host config

Minimal flake.nix for a host that runs hive-c0re:

{
  inputs = {
    nixpkgs.url = "github:NixOS/nixpkgs/nixos-25.11";
    hyperhive.url = "git+https://git.berlin.ccc.de/vinzenz/hyperhive";
  };

  outputs = { nixpkgs, hyperhive, ... }: {
    nixosConfigurations.my-host = nixpkgs.lib.nixosSystem {
      system = "x86_64-linux";
      modules = [
        hyperhive.nixosModules.hive-c0re
        ({ ... }: {
          services.hive-c0re.enable = true;
          # Free-text operator pronouns — defaults to "she/her", threaded
          # through to every agent's system prompt as HIVE_OPERATOR_PRONOUNS
          # so claude refers to you naturally in third person.
          # services.hive-c0re.operatorPronouns = "they/them";

          # ... rest of your host config (hardware, networking, users, …)
          system.stateVersion = "25.11";
        })
      ];
    };
  };
}

hive-c0re will then:

• open its admin socket at /run/hyperhive/host.sock + dashboard on :7000,
• auto-create the manager container (hm1nd) if missing,
• auto-rebuild any managed container whose hyperhive rev is stale.

claude-code is unfree; hyperhive whitelists it for itself (scoped: only claude-code, nothing else) inside the claude-unstable overlay and harness-base.nix. Per-agent containers evaluate their own nixpkgs instance so the operator's host-level allowUnfree doesn't propagate in — the predicate has to live inline. Nothing to set on the operator side.

Build / deploy

# inside the repo (devshell first; no global cargo)
nix develop -c cargo check
nix develop -c cargo clippy --workspace --all-targets -- -D warnings

# evaluate everything (rust+nix+toml fmt + clippy)
nix flake check

# deploy to a host that imports `hyperhive.nixosModules.hive-c0re`
cd ~/Repos/<nixos-config-repo>
nix flake update --update-input hyperhive
sudo nixos-rebuild switch --flake .#<host>

No overlays on the host's pkgs — the module pulls hive-c0re's package straight from hyperhive.packages.<system>.default. Just import the module and the service is wired up.