Orazaka

The AI that never leaves home.

Orazaka is the open-source sovereign AI platform by Krizaka (Montréal) — chat, documents, images, video and agents on your own machines — and this is the workspace that assembles it. Website: dev.orazaka.com.

Website · Documentation · Krizaka · Krizaka on GitHub

A multimodal AI orchestration engine — chat, RAG, agents, image, audio, video — where every request crosses a deterministic interceptor pipeline to the best local model. Your documents, your models, your network: nothing leaves it. Law 25 and GDPR by design.

Unfiltered models, your rules. Hosted assistants apply their vendor's refusal policy to every answer. Orazaka serves open-weight models without that layer — Qwen 2.5, Dolphin Llama 3 and Gemma 3, abliterated or fine-tuned without refusals (infra/migrations/2026-10-11-uncensored-models.sql) — so a security team, a clinician, a reporter or a novelist gets a straight answer, on your own machines. The policy is the operator's: a pack declares what it refuses (scopeGuard) and the crisis text it answers (safety), the engine enforces both before any model runs, and every turn is accounted for. "Uncensored" means not over-refusing legitimate work; it never covers content sexualising minors, explicit sexual content, real persons, weapons or bio-chemical procedures, working malware, or anything that enables serious harm. See the answers.

This repository is the entry point of the Orazaka platform. It holds the governance contract (AGENTS.md), the workspace manifest (orazaka.workspace.json), the local infrastructure (infra/), the hermetic end-to-end suite (orazaka-end2end/) and the architecture documentation (docs/). Every component lives in its own repository so that any Krizaka application can pick only what it needs.

Repositories

Orazaka foundation

Repository	What it is
orazaka-build	Parent POM of every Orazaka JVM repository: inherits krizaka-parent, imports krizaka-bom 0.2.0 and the Spring AI BOM, and carries the Orazaka governance tests.
orazaka-contracts	Tier-1 platform contracts shared by every Orazaka service: the job plane (jobs-api) and the application-persistence ports (persistence-app-api). Pure interfaces and records, zero implementation.
orazaka-edge	Transparent HTTP edge in front of every Orazaka service: routing, API-key → JWT exchange, CORS and tracing. Spring MVC + virtual threads.
orazaka-ui-kit	@krizaka/orazaka-shared (TypeScript types, Zod schemas, design tokens) and @krizaka/orazaka-design-system (React components, Tailwind preset, theme, icon registry) for Next.js and React Native apps.

Krizaka building blocks — reusable by any application

Released on Maven Central (com.krizaka, BOM 0.2.0) and cloned under krizaka/ so the reactor builds them from source. Documentation: krizaka.com/docs/java.

Repository	What it is
krizaka-build	Parent POM, krizaka-bom and the test kit (ArchUnit rules, Testcontainers helpers) of every Krizaka JVM repository.
krizaka-platform-kit	Web (problem details, correlation id), security (session JWT, SERVICE tokens), messaging (outbox, deduplication, retry / DLQ), observability — and their Spring Boot starters.
krizaka-users	Reusable user management: registration, e-mail verification, login (password + Google/GitHub OAuth), forgot/reset password, profile & preferences, API keys, RBAC and JWT issuance.
krizaka-notifications	Channel-based notification delivery: e-mail (SMTP), SMS (Twilio) and webhooks behind one DeliveryClient port, driven by platform events or explicit requests over AMQP.
krizaka-billing	Credits, wallets, plans, subscriptions, price book and metering (hold → settle → release), with its contract (billing-api) and a typed HTTP client (billing-client).

Orazaka AI engine

Repository	What it is
orazaka-studio	Studio & pack marketplace: pack catalogue, studio blueprints, installations and saga-driven runs, with its contract (studio-api) and HTTP client (studio-client).
orazaka-ai-engine	The Orazaka cognitive SDK: AiClient & provider mesh (core), the interceptor pipeline, tools (MCP · RAG · sandbox), use-case orchestration (business), application persistence and asset encryption.
orazaka-conversation-service	Interactive ingress of the Orazaka engine: chat (SSE streaming), intents, models, pipeline, MCP and job APIs — translates transport into Intentions, never business logic.
orazaka-job-service	Asynchronous job executor of the Orazaka engine: drains the interactive and batch lanes, owns the capability registry and the worker registry.
orazaka-knowledge-service	Knowledge & RAG retrieval service (pgvector) with asynchronous indexing.
orazaka-automation-service	Connector automation (Jira, Slack, WhatsApp, Messenger, CLI agents) with Quartz scheduling and execution telemetry.
orazaka-worker-media	Native (Metal) Python worker for image/video generation and media composition, speaking the Orazaka AMQP worker protocol.

Applications & content

Repository	What it is
orazaka-web-client	Next.js (App Router) client of the Orazaka platform with its mandatory BFF.
orazaka-web-admin	Retired — the operator console is served by orazaka-web-client on admin.orazaka.com (host routing). The repository is archived and deleted on 2026-10-23.
orazaka-mobile-client	Expo / React Native client of the Orazaka platform (talks to the edge with Bearer tokens).
orazaka-cli	The orazaka developer CLI: install, start, dev, demo seed, test, docs and pack tooling for the Orazaka workspace, with an offline SQLite queue.
orazaka-packs	Reference packs for the Orazaka Studio marketplace (document validation, media, prospection, real-estate, wellbeing…) and the pack manifest schema.

Dependency graph

graph LR
  krizaka_build --> krizaka_platform_kit
  krizaka_build --> krizaka_users
  krizaka_platform_kit --> krizaka_users
  krizaka_build --> krizaka_notifications
  krizaka_platform_kit --> krizaka_notifications
  krizaka_build --> krizaka_billing
  krizaka_platform_kit --> krizaka_billing
  krizaka_build --> orazaka_build
  orazaka_build --> orazaka_contracts
  orazaka_build --> orazaka_studio
  orazaka_contracts --> orazaka_studio
  krizaka_billing --> orazaka_studio
  orazaka_build --> orazaka_ai_engine
  orazaka_contracts --> orazaka_ai_engine
  krizaka_billing --> orazaka_ai_engine
  orazaka_studio --> orazaka_ai_engine
  orazaka_build --> orazaka_conversation_service
  orazaka_contracts --> orazaka_conversation_service
  krizaka_users --> orazaka_conversation_service
  krizaka_billing --> orazaka_conversation_service
  orazaka_ai_engine --> orazaka_conversation_service
  orazaka_build --> orazaka_job_service
  orazaka_contracts --> orazaka_job_service
  krizaka_users --> orazaka_job_service
  krizaka_billing --> orazaka_job_service
  orazaka_ai_engine --> orazaka_job_service
  orazaka_build --> orazaka_knowledge_service
  orazaka_build --> orazaka_automation_service
  krizaka_billing --> orazaka_automation_service
  orazaka_build --> orazaka_edge
  orazaka_ui_kit --> orazaka_web_client
  orazaka_ui_kit --> orazaka_web_admin
  orazaka_ui_kit --> orazaka_mobile_client

Rules (AGENTS.md §2): a repository depends only on repositories before it in orazaka.workspace.json (no cycle); a service depends on another context's contract (*-api) or client, never on its implementation.

Run Orazaka on your own infrastructure

On a Mac (Apple Silicon) or a Linux server with an NVIDIA GPU — your data never leaves your machine, no telemetry, offline once the models are downloaded. Requirements, the 3 steps per target, update, backup and uninstall: docs/SELF_HOSTING.md. How it is packaged and signed: docs/DELIVERY.md. Security reports: SECURITY.md.

Quick start

Requirements: git, JDK 21, Node.js 22+, Docker, Python 3.11+ (media worker), macOS recommended for native inference.

npx orazaka install                    # clones this workspace and every repository into ./orazaka, verifies the tools, configures it
cd orazaka
./mvnw install                         # builds and tests every JVM repository in one reactor
(cd orazaka-apps/ui && npm install)    # links the cloned UI repositories
npx orazaka start                      # local infrastructure: Docker middleware and AI engines
npx orazaka dev                        # the whole stack: every service, Web, Admin and Mobile
npx orazaka demo seed                  # second terminal, stack up: Eric, the demo persona (account, plan, packs, Studios)

Sign in on the web client as Eric — the account orazaka demo seed prints — and replay the demo above on your own machine. npx orazaka is @krizaka/orazaka-cli on npm (orazaka is its short name; npx @krizaka/orazaka-cli … is the canonical form). By hand instead of install: git clone https://github.com/krizaka/orazaka.git && cd orazaka && node scripts/workspace.mjs clone.

node scripts/workspace.mjs status shows the branch and state of every repository; node scripts/workspace.mjs pull updates them all.

Follow Studio runs without polling

Studio runs expose per-run and per-key Server-Sent Events streams, and signed run.completed / run.failed webhooks for server-to-server integrations. Webhook endpoints are managed in Settings → Webhooks; delivery is at-least-once, so receivers should deduplicate on webhook-id. See the Studio guide for event resumption, signatures, retries, and the delivery log.

Grounded research (proposed)

The engine looking things up, reading the source and citing it, instead of answering from memory: ADR-087 and its design (proposed, not built yet).

The compliance checks built from the law, on top of it — identify the document, read the dated articles, propose rules, promote them to dated rulesets: ADR-089 (proposed, not built yet).

Workspace layout

orazaka/                               ← this repository
├── AGENTS.md · orazaka.workspace.json · pom.xml (reactor)
├── infra/        docker-compose, initdb (reset + dev fixtures), migrations
├── docs/         architecture, ADRs, generated references
├── orazaka-end2end/                   hermetic E2E
├── orazaka-libs/                      ← cloned: orazaka-build, orazaka-contracts, orazaka-ai-engine
├── krizaka/                           ← cloned: krizaka-build, krizaka-platform-kit, krizaka-users, krizaka-notifications, krizaka-billing
├── orazaka-apps/services/             ← cloned: orazaka-studio, conversation, job, knowledge, automation, edge
├── orazaka-apps/workers/              ← cloned: orazaka-worker-media
├── orazaka-apps/ui/                   ← npm workspace root; cloned: orazaka-ui-kit, web, mobile, cli
└── orazaka-packs/                     ← cloned: orazaka-packs

Cloned directories are ignored by this repository's git; each is its own repository.

Building a new application from Orazaka components

1. Import com.krizaka:krizaka-bom 0.2.0 from Maven Central and add the starters you need (krizaka-spring-boot-starter-web, -security, -rabbitmq, -observability) — or inherit com.krizaka.orazaka:orazaka-parent (from orazaka-build) for the full Orazaka stack and quality gates.
2. Run the services you need — e.g. krizaka-users for registration / login / forgot password / profile, krizaka-notifications for e-mail / SMS / webhook delivery, krizaka-billing for credits and subscriptions, orazaka-edge in front of them.
3. Depend only on their contracts (krizaka-users-api, krizaka-notifications-api, krizaka-billing-api) or clients.
4. Build the UI with @krizaka/ui and the Orazaka identity, @krizaka/orazaka-design-system.

Consuming packages

• Krizaka building blocks (com.krizaka) are on Maven Central — no repository to declare. Import the BOM: com.krizaka:krizaka-bom:0.2.0 (all artifacts).
• npm packages (@krizaka/ui, @krizaka/orazaka-design-system, @krizaka/orazaka-shared, …) are on the public npm registry — no .npmrc, no token.
• Orazaka JVM artifacts (com.krizaka.orazaka) are still published to GitHub Packages on a v* tag. Maven, in ~/.m2/settings.xml:

<servers>
  <server><id>github</id><username>YOUR_GITHUB_USER</username><password>${env.GITHUB_TOKEN}</password></server>
</servers>
<!-- and a <repository> with id "github" per Orazaka repository you consume, e.g.
     https://maven.pkg.github.com/krizaka/orazaka-build -->

License

Apache License 2.0 — see LICENSE and NOTICE.

---

A Krizaka product · krizaka.com · Apache-2.0