Feature Guide

HivemindOS is a control room for local agent fleets.

The feature surface follows the work an operator actually does: connect machines, configure agents, run work, preserve memory, manage money rails, and keep the local system healthy.

Control Room

The product starts with machines and agents. Fleet tells you what is online, which collectors are reachable, what runtimes and apps are available, and whether a machine is drifting from the expected setup. Agents and Chat turn those runtimes into configured workers with models, env, wallets, vault context, file attachments, and session history.

Looped Work

Looped work is where operator intent turns into agent execution. A loop contract can start from chat, Scheduler, Queen Bee, Evo, a company launch, or the Work Board. The board remains one access path: it captures rough ideas, promotes ready tasks, tracks claimed work, stores comments and run records, and turns finished output into deliverables. HivemindOS evaluates managed completions so agents do not get credit just for saying done. Zero Human Companies add a company cockpit with two autonomy choices: the default HivemindOS crew engine uses the Work Board and private learning loops, while the optional AEON engine runs one saved background skill and records accepted handoffs in Company Runs. Scheduler adds repeated background work. Chat can launch /swarm [number] agent-team passes for parallel role-specific analysis, or /swarm-goal <build request> to rewrite a loose build request and hand it to Queen Bee for parallel agent execution. Swarm and MiroShark handle rehearsal, /swarm-sim simulations, and heavier analysis workflows.

Code Proof

Agent work gets messy fast if the only record is a chat transcript or a commit message. HivemindOS keeps the private trail: which task was opened, which project it belonged to, which machine handled it, and which agent touched it. GitLawb adds the part that code needs, signed provenance.

The goal is not to make every user manage a code node on day one. The normal path is quieter than that. HivemindOS can be Code Proof ready during setup, tasks can attach to projects, and linked projects can carry GitLawb proof metadata. If a project later needs local repo hosting, the deeper GitLawb node path is there.

That gives the Work board a better answer than “an agent changed something.” It can point to the project, the task, the machine, and the proof trail behind the code.

Shared Brain

The shared brain is a normal Obsidian vault, not a proprietary database. HivemindOS writes durable state into that vault when available: Kanban records, notifications, scheduled runs, wallet records, shared skills, service notes, and reviewed outputs. Agent Memory supports entity-linked recall, canonical memory heads, temporal history, and soft retrieval telemetry; high-volume operational events stay in a separate bounded local journal and only become durable knowledge through review. QMD, GBrain, Neo4j, Syntho, and Trading Brain stay optional service layers around the vault.

Economy And Integrations

Wallet and token features are explicit rails, not a background permission pool. Agent wallets handle controlled Base, Robinhood Chain, and Solana balances plus x402 paid requests, including HivemindOS Models calls paid directly from a local wallet without user API keys. The crypto capability router lets agents ask for intents such as paid API, private transfer, tokenized stock trade, or Bankr trading, then select or prepare the configured rail without executing spending itself. Hive Compute is an experimental marketplace inference and spare-GPU hosting path. Honey is one cumulative record of verified contribution and bounded recognition, Hivemind Cloud credits pay for managed service usage, and HIVE remains an optional community/payment rail. Integrations connect the control room to outside systems without making those systems own local state.

Native And Background Surfaces

The browser and native app share the same Next.js UI. Tauri adds local filesystem and status helpers on This Mac, while AEON and Hermes keep their runtime-specific setup and background work docs separate.

Feature Map

flowchart TD
  Fleet["Fleet"] --> Runtimes["Agents and runtimes"]
  Fleet --> Apps["My Apps and API services"]
  Runtimes --> Chat["Chat"]
  Runtimes --> Phone["Phone calls"]
  Chat --> Work["Work board"]
  Chat --> XRay["Context X-Ray"]
  Work --> Scheduler["Scheduler"]
  Work --> LoopEngineering["Loop engineering"]
  LoopEngineering --> Evaluations["Agent evaluations"]
  Work --> VisualPlans["Visual plans and recaps"]
  Work --> Deliverables["Deliverables"]
  Work --> CodeProof["GitLawb Code Proof"]
  Work --> History["Work history"]
  Vault["Brain and vault"] --> Skills["Shared skills"]
  Vault --> ReviewQueue["Review queue"]
  Vault --> GBrain["GBrain"]
  Vault --> Syntho["Syntho / Synthesis"]
  Vault --> Notifications["Notifications"]
  Vault --> Env["Shared env"]
  Runtimes --> OpenClaw["OpenClaw"]
  Runtimes --> Evo["Evo optimization runs"]
  Runtimes --> MiroShark["MiroShark"]
  Runtimes --> AEON["AEON"]
  Work --> Companies["Zero Human Companies"]
  Companies --> AEON
  MiroShark --> Swarm["Swarm theater"]
  Runtimes --> Wallets["Wallets and tokens"]
  Wallets --> HiveCompute["Hive Compute payments"]
  Wallets --> TokenRails["Base / Solana / USDC"]
  Wallets --> Bankr["Bankr HIVE"]
  CodeProof --> Projects["Project registry"]