A git-backed operating system for running a business with AI agents. Plain Markdown and YAML, readable by any file-reading agent, owned by you.

The Infinite Brain is a knowledge OS: it makes what your business knows, decides, and does reliably retrievable and safely executable by AI agents, today and after the tools change. Knowledge lives in namespaces with an explicit promotion path to operator-approved canon. Work lives in projects with typed entities (commands, agents, skills, rules, workflows, tools). A contract layer (_system/) plus a validator keeps the whole graph honest. Your AI coding agent (Claude Code, Codex, or any file-reading agent) is the runtime.

No database, no server, no vendor lock-in. If you can read this repo, so can your agents.

git clone <this-repo> my-brain && cd my-brain
bash _system/validate.sh     # the contract holds: zero errors on a fresh clone
claude                       # or your agent of choice; CLAUDE.md orients it

Then say to your agent:

Read START-HERE.md and give me the tour.

Or open the folder as an Obsidian vault (config ships in .obsidian/) and read START-HERE.md yourself. The full walkthrough is docs/getting-started.md.

• The contract layer. _system/ holds the schemas, rules, registries, and validate.sh: what must be true in this repo and how it is checked.
• The doctrine. knowledge/ai-architecture/ is the full reference architecture: the control spine, the namespace model, canon versus synthesis, retrieval doctrine, surface boundaries, and the agent-authority limits. It is the "why" behind every folder here.
• The OODA orientation set. How the whole OS reads as John Boyd's real OODA web: the bridge, the feedback-plane spec, and a router index in knowledge/ai-architecture/synthesis/, the wager-ledger design that closes the Act-to-Orient arrow, and an interactive visual explainer (docs/ooda-infinite-brain-map.html).
• A complete worked example. A fictional candle studio (Emberline) threads through one example of every entity type, all cross-linked: a command, an agent, a skill, a rule, two workflows (agentic and deterministic), a tool, a tiny knowledge namespace, a data pointer, a memory, an output, a filled project, and an assembled department.
• Builders. Skills and workflows that scaffold new namespaces, departments, projects, agents, and knowledge bases, plus an onboarding interview (docs/onboard-business.md) that maps your business onto the architecture.
• Eight profile references. knowledge/_examples/ shows the eight namespace profiles (doctrine, data-system, design-system, content-strategy, tool-contract, and more) as copyable scaffolds.
• Provenance. PROVENANCE.yml at the repo root records the exact source commit, export date, and pipeline version this release derives from, machine-readable.

_system/           The operative contract: schemas, rules, registries, validate.sh
knowledge/         Namespace-first knowledge graph (the doctrine and your domains)
entities/          Canonical executable entities: commands, agents, skills, rules
.claude/ .codex/   Runtime adapter shims (regenerate with sync-adapters.sh)
workflows/         Agentic reasoning pipelines
automations/n8n/   Deterministic workflows (JSON plus a brain-record companion)
tools/             Pointer nodes over bounded capabilities
departments/       Assemblies over the entities: one folder per operating lane
projects/          One PLAN.md per project, with inline tasks
intake/            Inbound flow: source captures, routing, processed receipts
data/              Pointers to where numbers live (never the numbers)
memory/            Reviewed learnings
outputs/           Produced artifacts with lineage
sessions/          The audit trail of AI work sessions
swarms/            Multi-agent sprint packages
docs/              Setup, retrieval, and onboarding docs

1. knowledge/emberline-studio/canon/brand-essentials.md: a canon node, the studio's source of truth. Note the verification fields and the changelog.
2. entities/rules/studio-brand-voice.md: a rule derived from that canon.
3. entities/skills/write-product-description.md: a skill that applies the rule.
4. entities/agents/studio-inbox-triage.md: an agent that uses both and escalates what it must not decide.
5. workflows/weekly-studio-review.md: the weekly loop reading data/orders-ledger.md (a pointer, never live numbers) and memory/photograph-before-listing.md (a lesson).
6. outputs/2026-06-05-spring-collection-brief.md: what the loop produced, with lineage.
7. projects/_example/PLAN.md: the work container that ties it together.
8. departments/example-studio-ops/INDEX.md: the assembly of all of the above.

Every file is under two minutes' reading. Follow the edges in the frontmatter; that is the graph.

• Every node-bearing file carries typed YAML frontmatter; bash _system/validate.sh must exit 0.
• Canon is operator-approved, always. Agents draft; you sign.
• The repo never stores live numbers, live queues, or secrets: pointers only.
• Sessions that touch the repo are registered, logged, and closed out in sessions/.

Shipped doctrine occasionally refers to Paperclip, the task runtime of the deployment this starter derives from: treat it as a placeholder for whatever runtime you adopt. Nothing here requires it.

The starter still self-describes as architecture v3.1; this release extends the content, not the retrieval or ontology spec. Headlines:

• The OODA orientation lens and the feedback-plane design. Canon now frames the OS with Orient as the dominant center (core-doctrine sections 14 and 15.2, department-model section 11), backed by the shipped synthesis set: the Boyd bridge, the Act-to-Orient feedback-plane spec, the wager-ledger decision and operative contract (_system/wager-ledger-rules.md), and the interactive visual map in docs/. The wager ledger is a ratified design, not a running system; it ships as doctrine you can build.
• The reflexive-brain-topology enterprise standard. How a company organizes repos once it runs the brain internally: shared parent, individual brains (one per person), department brains, one company brain (knowledge/ai-architecture/pillars/reflexive-brain-topology.md), plus the trust-boundary graduation decision and playbook for when a department earns its own repo.
• repo_kind and brain_tier registry fields. _system/repo-registry-rules.md and the registry template now classify every repo as brain, app, or mixed, and every brain as individual, department, or company.
• The asset-reference layer. _system/asset-reference-schema.md and _system/asset-registry-rules.md: binary assets stay out of git, references and metadata stay in, the same discipline as secrets.
• Department operations. The department-web canon (capture, convert, build, operate, feed back), the ambient capture rule, the onboarding and operating guides, the operations-readiness gate, and the daily-update workflow.
• Intake playbooks. Processing procedures for the email and Slack source lanes (intake/playbooks/), alongside the existing X, YouTube, web, repo, and research lanes. Capture connectors are not bundled: you wire your own capture into intake/sources/, and a connector suite is the planned next release.
• Machine-readable provenance. PROVENANCE.yml replaces prose-only lineage: source commit, source dirty-path count, export date, pipeline version, spec version.
• The companion harness repo. The shared-parent tier of the topology is now a real, runnable repo: infinite-brain-harness is a thin versioned root that holds this brain and its siblings, with the registry and orientation layer for running multiple brains side by side.

Run the onboarding interview (docs/onboard-business.md), or go manual: build your first namespace with entities/skills/build-namespace.md, assemble your first department from departments/_template/, fill in knowledge/personal-operator/pillars/operator-profile.md, and retire the candle studio when you no longer need the training wheels.

Growing past one repo? Put your brains under the companion infinite-brain-harness root: it carries the shared-parent orientation and repo registry from the reflexive-brain-topology standard, so a second brain is a registry entry, not a redesign.

MIT. See LICENSE. Contributions welcome: see CONTRIBUTING.md.