What is FractalMind?

FractalMind AI is open-source infrastructure for running AI agent teams as a governed operating system.

The Problem

Most agent tooling stops at one layer:

• chat interfaces move messages
• agent frameworks manage one worker at a time
• team frameworks coordinate a local swarm
• protocol projects define trust surfaces

But real delivery needs the whole loop: shared state, priorities, governance, execution, evidence, and learning. Without that, multi-agent work becomes fragile, opaque, and dependent on heroic manual coordination.

The Current Answer

FractalMind combines a local governed operating loop with three remote-control planes:

• Control / Authority Plane — SUI contracts define target-scoped capability, delegation, revocation, use, and budget bounds.
• P2P Data / Execution Plane — target envd verifies signed intents, protects replay state, and invokes typed local adapters.
• Application Plane — Agent Console, envd-desktop, channel adapters, and automation clients create requests and display results.
• Local operating loop — heartbeat, structured memory, candidate OKRs, governance, evidence, and outcomes coordinate delivery.

The control loop is explicit:

signal -> memory -> candidate OKR -> governance -> execution -> outcome -> evolution

How It Works Today

The three-plane architecture is the migration target. Existing repositories still contain compatibility paths, so public docs distinguish current implementation from target behavior.

For remote privileged actions:

capability -> signed intent -> P2P/relay -> target envd verify
           -> durable reservation -> local adapter -> durable result

Relay, coordinator, channel identity, and bearer tokens are not authority sources. Raw logs, media, input, files, and frequent heartbeats remain off-chain.

• The daily running core lives in oh-my-code
• Shared candidate OKRs land in fractalmind-okrs
• Public trust surfaces still matter, but they are no longer the only story

Who Is This For?

• AI builders who need more than a chat bot or single-agent wrapper
• Teams operating multiple agents with real delivery pressure
• Researchers exploring governed autonomy and long-horizon coordination
• Organizations that want auditability, human–AI co-creation, and repeatable execution

Current State

As of the current public org state:

• FractalMind AI spans 18 public repositories
• The heartbeat-driven OS loop is running in oh-my-code
• Candidate OKR sync is active through fractalmind-okrs
• The existing protocol package remains live on SUI Testnet for identity, tasks, and governance
• Remote capability and target-enforcement work remains a tracked migration until reviewed, merged, and published
• Current migration status is tracked in fractalmind-ai/.github#6

See the Architecture Overview and Roadmap for the latest framing.