Javis OS
Javis OS is an Agentic AI that assists entrepreneurs in business tasks, helping them grow their work.
AI translation
One living graph runs the business. Built by an AI agent org under human ratification; designed for AI-assisted teams and, tomorrow, their customers' agents.
BusinessGo · JavaScriptGitHub @operatinggraphOnlineOwner unconfirmed
Have you tried it? One tap is enough. No sign-in needed.
No updates detected yet. GitHub releases will appear here.
Owners of public repositories under a personal GitHub account can sign in to verify ownership and manage the introduction themselves. No administrator approval is needed.
Sign in, generate a verification file, and commit it to this repository’s default branch within 30 minutes. Files on other branches or forks do not count. Then check the file below to manage the project without administrator approval. Keep the file to retain your access.
If needed, you will go to sign-in. After signing in, return here and press your chosen verification button again.
Run /nomorevibe verify in your project folder to verify ownership and manage the introduction yourself.
Verify in one minuteAsk about other verification options
The business is a graph — run it on one. A graph-native, agent-native platform on NATS where every capability installs like a package and evolves while the system runs, safe for humans and AI agents to change together.
Taken from the repository README
Loading the full README. The saved excerpt is shown below.
A graph-native, agent-native application platform built on NATS. ⚠️ Work in progress. Lattice is an active, in-development project. See Project status for what's implemented today. Website: operatinggraph.com — what Lattice is, why it exists, and the story of the AI agent organization that builds it (the About page carries the long-form reading). This README is the technical front door: the architecture, the status, and how to run it.Read the full README ↗
A graph-native, agent-native application platform built on NATS.
⚠️ Work in progress. Lattice is an active, in-development project. See Project status for what's implemented today.
Website: operatinggraph.com — what Lattice is, why it exists, and the story of the AI agent organization that builds it (the About page carries the long-form reading). This README is the technical front door: the architecture, the status, and how to run it.
Lattice runs an entire business domain as one addressable graph — the VAL model: vertices (entities), aspects (their data, each independently addressable), and links (typed relationships that double as the ReBAC authorization graph). Types, operations, schemas, and permissions are themselves in the graph, so the system can describe itself to any actor — including AI agents, which discover what they may do by walking from their own identity instead of depending on a hand-written SDK.
A few invariants carry the whole design:
core-operations stream; the
Processor is the sole writer to Core KV — a 9-step commit pipeline with schema-aware
Starlark validation, capability authorization, atomic batch commit, and a transactional event
outbox. There is no side door for state mutation.Lattice is a small set of cooperating components, each with a living reference page:
| Component | Role |
|---|---|
| Processor | The sole authorized writer — a 9-step commit pipeline that runs Starlark over Core KV, with atomic batch commit and a transactional event outbox |
| Refractor | The read side — continuous openCypher lens projections (Postgres / NATS KV), the security-critical Capability Lens, and CDC consumers |
| Substrate | NATS / KV / NanoID primitives — key shapes, atomic batch, durable CDC consumers |
| Capability Packages | Installable bundles (identity, RBAC, domain logic) added through the InstallPackage kernel op — the kernel stays minimal |
| Loom | The procedure engine — deterministic, idempotent, linear flows (the "executive") |
| Weaver | The convergence engine — drives a declared target state toward convergence (the "visionary") |
| Augur | The Weaver's AI-assisted reasoning tier — escalates a gap the deterministic planner can't solve to a model, records the answer as a validated, human-reviewed proposal, then dispatches it back through the Weaver (the Processor stays the sole writer) |
| Bridge | The external-I/O egress — the one component that makes outbound calls to external systems, idempotently, via a durable events.external.> consumer and a pluggable adapter registry |
The exact wire shapes, key patterns, and behavioral rules are pinned in the data contracts under
docs/contracts/.
Lattice is deliberately developed by AI agents — the code, the tests, the contracts, and the
documentation are agent-written, with one human principal as architect and supervisor: setting the
vision, freezing the data contracts, and ratifying every design and frozen-contract change. Since
Phase 3 the loop runs autonomously on a schedule — agents survey demand, design, build behind the
full gate suite, and exercise the shipped apps to file the next round of demand. The paper trail
is in-repo: the live boards under
_bmad-output/planning-artifacts/backlog.md,
the design docs under _bmad-output/implementation-artifacts/, and the role charters in
agents/README.md. The longer story is on the
website's About page.
This is the one place that distinguishes what's built from what's designed.
| Phase | Scope | State |
|---|---|---|
| Phase 1 | Trustworthy core: substrate, Processor write path, Refractor lens projections, identity/RBAC packages, Capability-Lens authorization, the Hello Lattice reference slice | ✅ Implemented + tested (CI-gated) |
| Phase 1.5 | Hardening: kernel minimization, package installs routed through the Processor, contract conformance suite, transactional event outbox | ✅ Complete |
| Phase 2 | Orchestration: Loom (procedures) + Weaver (convergence) + the external-I/O bridge + the Loftspace lease-application reference vertical | ✅ Complete (CI-gated) |
| Phase 3 | Driven by the autonomous agentic flywheel (see above). Shipped: the Loupe operator console and four vertical front-ends (LoftSpace, Clinic, Café, Wellness) end-to-end on the real stack; the Gateway trust boundary with real-actor write authorization and operator login (capability authorization is the only mode); Vault crypto-shredding with audited reveal; the Chronicler history materializer; the Augur propose → review → dispatch loop; Semantic Contracts as a sanctioned pattern + reference package; the first production Personal Lenses and the Edge Lattice reference node (offline-first read/write, per-identity connection confinement). In build: the Edge in-browser node (wasm) and the Facet discovery-driven personal client. Designed + deferred: multi-cell sharding, HA clustering. | 🏗️ Continuous |
docs/architecture-overview.md — full platform architecture diagram (as designed, all phases)docs/ — the documentation map (contracts, components, observability, operations)docs/contracts/ — the data contracts (source of truth for wire shapes)docs/components/ — living per-component reference pagesdocs/hello-lattice.md — the 60-minute end-to-end tutorialThe longer-form vision (the Lattice Manifest and System Spec) lives in a separate design vault;
the architecture of record is in docs/.
Prerequisites: Go 1.26+, Docker + Docker Compose, make.
# Minimal kernel — NATS + Postgres, primordial bootstrap, Processor + Refractor
make up
# Confirm everything is healthy
lattice health gates
Expected output:
health.gates.phase1.gate1 passed: true
health.gates.phase1.gate2 passed: true
health.gates.phase1.gate3 passed: true
health.gates.phase1.gate4 passed: true
# Full platform — adds the Loom/Weaver/Bridge engines, the core capability
# packages (identity, RBAC, privacy, objects), the Gateway (:8080), and the
# Loupe operator console → http://127.0.0.1:7777
make up-full
Each target brings the full platform up (reusing a running stack), installs the vertical's capability packages onto the live graph, and starts its front-end:
make up-loftspace # LoftSpace residential leasing → http://127.0.0.1:7788
make up-clinic # Clinic appointment booking → http://127.0.0.1:7799
make up-cafe # Café house-tab ledger → http://127.0.0.1:7801
make up-wellness # Wellness classes & rosters → http://127.0.0.1:7802
The verticals compose: install more than one and they share the graph — a completed Clinic visit can book a Wellness class with no integration code written.
Then walk the full vertical slice by hand — define a type, create entities, project to Postgres, drive it with an AI agent, and roll a schema change back — in the Hello Lattice tutorial.
# Build all binaries
make build
# Run all unit + integration tests (serialised for embedded NATS stability)
make test
# Lint
golangci-lint run ./...
# Go vet (ANTLR-generated files excluded)
make vet
# Gate tests
make verify-kernel
make test-rollback
make lint-conventions
The security proofs live in each mechanism's own colocated tests — internal/bypass (the
outcome-level residual), internal/natsperm, the Processor's Starlark gates, the Refractor's RLS
tests, the Gateway auth tests — and all run under make test.
Packages upgrade in place on a running stack — no make down. After editing a
package's DDL, lens, or permissions:
# Diff-apply one edited package (create/update/tombstone in one atomic op)
make reinstall-package PKG=packages/clinic-domain
# Refresh a whole vertical: diff-apply its packages + rebuild/restart its FE binary
make refresh-clinic # or: refresh-loftspace / refresh-cafe / refresh-wellness
# Preview the delta without submitting
lattice-pkg install --dry-run packages/clinic-domain
A newly-added lens/role/op hot-activates live too, same as an edit — the Refractor's CDC watch and the Processor's DDL cache both react to any commit, not just updates. Only a change to the primordial kernel seed needs a fresh bootstrap. See Capability Packages → Upgrade.
Lattice is source-available, not open source: the code is public to read, run locally, and evaluate; all other rights are reserved — see LICENSE. For commercial licensing or partnership inquiries: [email protected].
| The Chronicler | The event-ledger materializer — tails the platform's event and intent-ledger streams into append-only history read models: the durable audit counterpart to Refractor's present-state projections (a separate binary by charter) |
| Object-store-manager | The byte-janitor of the off-graph blob plane — reclaims object-store bytes when their vertex is tombstoned or crash-orphaned, and cascades owner-tombstones to detach dangling object links |
| Gateway | The edge trust boundary — verifies an external actor's IdP-signed JWT, stamps the verified identity onto every operation, and bounds each actor's read view to the sub-graph its ReBAC links permit, closing actor impersonation at the edge |
| Vault | Per-identity key custody and crypto-shredding — encrypt-on-write / decrypt-on-read for sensitive aspects, and the irreversible ShredKey "right to be forgotten" primitive; a library embedded in the Processor and Refractor, with an async privacy-worker |
| Loupe | The internal operator console — browse Core KV, drive the engines' control planes, submit DDL-driven ops, install packages, upload blobs; a trusted single-identity, loopback-bound inspector (the one app allowed to read Core KV directly) |