Hive Superbrain

The Hive Superbrain is the shared intelligence and operating layer behind HivemindOS. It gives your permitted agents a common memory, source library, skill shelf, work context, and setup—so the hive can continue what you started instead of rebuilding context and configuration in every conversation.

Its durable core is a normal Obsidian vault. You can open it, search it, edit it, back it up, or sync it with tools you already trust. HivemindOS adds focused recall, reviewed memory, document intake, knowledge compilation, reusable skills, agent coordination, portable exports, and recovery tools around that human-readable foundation.

Shared Env sits beside the vault as the Superbrain’s credential layer. Secret values stay outside Obsidian, while safe key-name and set/missing references can live in the vault. That lets agents know which capabilities are ready without copying credentials into notes, prompts, or repositories.

More than memory

One brain for every way you work

The Hive Superbrain is not limited to a single chat window. HivemindOS-managed chats receive relevant memory before an agent starts. Raw runtime CLIs can use the same brain through hive-brain, which finds the local service and falls back to the local vault when the app is unavailable. Claude Code can receive the same recall through its installed prompt hook.

That means Codex, Claude, Hermes, Aeon, OpenClaw, Gemini, app-routed agents, and shell workflows can work from the same reviewed context when they have access. A current instruction still wins over an old memory, and workspace or machine policy can keep an agent out of context it should not receive.

The setup-once layer

Shared Env solves the part of multi-agent work that ordinary notes cannot: credentials and runtime configuration.

• Add a provider key once instead of pasting it into every agent, project, and machine.
• Let supported runtimes use the value at execution time without putting it in the conversation.
• Keep key names and set/missing status searchable in safe Operations/Secure/ references, while plaintext values remain outside Obsidian.
• Use the shared value as the fleet-wide default, while allowing a project or shell to override it for one specific job.
• When Hivemind Sync is enabled, reconcile additions and removals across trusted machines. Offline peers can catch up later, and tombstones keep an old peer from resurrecting a removed key.
• Restrict a machine’s Shared Env access to Allow, Ask, or Deny.

This is the Superbrain’s set it up once and stop babysitting every runtime capability: agents can discover that a provider is configured and use it when permitted, without learning or repeating the secret value themselves.

Open the Shared Env guide

How the Superbrain compounds

1. Capture

Drop in a file or folder, attach a source in chat, save a quick note, or let finished conversations mirror into the vault. HivemindOS keeps source provenance so later answers can distinguish evidence from interpretation.

2. Recall

For each request, the Superbrain checks the smallest useful layer first: concise reviewed memory, then relevant Markdown elsewhere in the vault, then focused Compiled Knowledge when it exists. It does not pour the entire vault into every prompt.

3. Refine

Important memories wait for Brain Review. Corrections create a new current version while preserving what came before. Repetitive run receipts stay in a separate bounded journal, so durable memory does not become a junk drawer.

4. Reuse

Reviewed knowledge can become a shared skill, a compiled topic wiki, a project decision, or a reusable source for the next agent. The result compounds across conversations instead of disappearing with one chat.

5. Coordinate

The same brain can connect work to shared projects, schedules, the Work Board, available machines, capable agents, and handoff receipts. Knowledge and execution stay linked without pretending that a note is a live runtime.

What you can ask it to do

• “What did we decide about the Atlas launch, and what changed since then?”
• “Read this research folder, preserve the sources, and build a reusable topic guide.”
• “Remember that weekly reports must be short and link to the source data.”
• “Turn this reviewed playbook into a skill every supported agent can use.”
• “Use the provider we already configured; do not ask me for the API key again.”
• “Continue this project on the best available machine and keep the handoff verifiable.”
• “Give this client a separate workspace so its notes, skills, and credentials cannot mix with ours.”
• “Export only the approved memory for this collaborator, with an expiry and passphrase.”

Built to move, split, and recover

• Readable source: durable knowledge remains ordinary Markdown and Obsidian files rather than being trapped in a hosted memory database.
• Separate workspaces: a client, team, bot, or experiment can have its own vault, Shared Env, skills, and brain-service state.
• Trusted-machine sync: vault content, credential delivery, and artifact handoffs use distinct paths so each can be enabled and governed separately.
• Portable subsets: scoped brain capsules can carry selected memories and Compiled Knowledge with checksums, optional encryption, expiry, embedded read-only search, and review-only import. OKF export provides an open bundle for outside agents and graph tools.
• Recoverable indexes: generated search state uses verified checkpoints and deltas, can fall back from damaged generations, and can be rebuilt from the human-readable source.
• Auditable repair: health checks report missing or stale pieces; vault doctor is read-only until you explicitly approve a fix, and fixes leave migration manifests.

What stays under your control

• Your vault is the editable source of truth for notes, memory, skills, projects, and Compiled Knowledge.
• A request to remember something becomes a Brain Review proposal before it changes durable memory.
• New reviewed truth can supersede an older memory without erasing the history.
• Raw imported content is treated as source material, not as instructions for an agent to obey.
• Plaintext secrets stay out of notes. The vault may record credential key names and status, not the values.
• Shared Env access can be controlled per machine, and project-local values override shared defaults.
• Workspaces can isolate knowledge, skills, configuration, and generated brain state.
• Optional search, graph, and synthesis services remain derived helpers; they do not replace the vault.
• The instruction you give now takes priority over recalled context that no longer applies.

Choose what you need

What happens when you ask a question

1. Shared Brain Memory checks reviewed facts, decisions, preferences, commitments, and other durable context.
2. If that layer is weak, full-vault search ranks relevant projects, conversations, sources, skills, and safe credential references.
3. Compiled Knowledge and the note graph can add focused, connected context when the task needs them.
4. The context index finds relevant capabilities and delivery paths without loading every tool definition or document.
5. The selected agent receives the useful context, applicable skills, and names-only setup status it is allowed to use.
6. If the task should change durable knowledge, the result goes through the appropriate review or authoritative work path instead of silently rewriting the brain.

You can ask about older context with phrases such as “what did we believe before the launch?” Corrected information remains part of the history, while current questions prefer the latest reviewed version.

flowchart TD
  User["User and agents"] --> Managed["HivemindOS-managed chats"]
  User --> RawAgents["Raw runtime CLIs"]
  Managed --> ContextIndex["Context and capability index"]
  Managed --> BrainMemory["Shared Brain Memory"]
  RawAgents --> HiveBrain["hive-brain CLI"]
  RawAgents --> ClaudeHook["Claude prompt hook"]
  HiveBrain --> BrainMemory
  ClaudeHook --> HiveBrain
  Drop["Brain Drop, chat attachments, and notes"] --> Vault["User-owned Obsidian vault"]
  BrainMemory --> Typed["Reviewed Agent Memory"]
  BrainMemory --> VaultSearch["Full-vault search"]
  Vault --> VaultSearch
  Vault --> Conversations["Conversation notes"]
  Vault --> Compiled["Compiled Knowledge and graph"]
  Vault --> Skills["Shared skill shelf"]
  Vault --> SecureRefs["Names-only secure references"]
  SharedEnv["Shared Env values outside the vault"] --> RuntimeProviders["Permitted runtimes and agents"]
  SecureRefs -. key name and status .-> SharedEnv
  Skills --> RuntimeProviders
  ContextIndex --> RuntimeProviders
  Vault --> QueenBee["Queen Bee, Work Board, and fleet coordination"]
  QueenBee --> Transfers["Verified machine handoffs"]
  Vault --> Capsules["Scoped capsules and OKF export"]
  Vault --> Generations["Verified index generations and recovery"]
  Vault --> QMD["QMD markdown search"]
  Vault --> Neo4j["Neo4j derived graph"]

~/Documents/Obsidian/hivemindos-vault

pnpm vault:doctor
pnpm vault:doctor -- --fix

More guides

• Vault Map
• Brain Services
• Memory Benchmarks
• Packaged Skills
• Shared Env
• Workspaces
• Sync And Health