Ordo

<p align="center"> <img src="images/ordo-logo.png" alt="Ordo Logo" width="180" /> </p> <h1 align="center">Ordo</h1> <p align="center"> <strong>The deterministic decision layer for AI agents — the LLM proposes, your rules dispose.</strong> </p> <p align="center"> A sub-microsecond, JIT-compiled rule engine in Rust. Give your AI agent guardrails you can <em>test</em>. </p> <p align="center"> <a href="https://ordo-engine.github.io/Ordo/"><img src="https://img.shields.io/badge/demo-playground-brightgreen" alt="Playground" /></a> <img src="https://img.shields.io/badge/rust-1.83%2B-orange?logo=rust" alt="Rust" /> <img src="https://img.shields.io/badge/license-MIT-blue" alt="License" /> <a href="https://www.npmjs.com/package/@ordo-engine/cli"><img src="https://img.shields.io/npm/v/@ordo-engine/cli?label=npm&color=cb3837" alt="npm" /></a> <a href="https://discord.gg/Y529FkArhh"><img src="https://img.shields.io/badge/discord-join-7289da?logo=discord&logoColor=white" alt="Discord" /></a> </p> <p align="center"> <img src="images/guard-demo.gif" alt="ordo guard blocking a destructive agent command — deny, ask, allow — with a reason" width="760" /> </p> --- An LLM is non-deterministic: ask it the same thing twice and you can get two answers. That's fine for drafting prose and dangerous the moment an agent touches money, data, or a shell. Ordo is the layer that decides — **allow / deny / ask** — deterministically, from rules that are versioned, tested, and auditable. The agent proposes the action; Ordo disposes. ## Guard your coding agent in 5 minutes The sharpest place to feel this today: put deterministic guardrails on a coding agent. `ordo guard` runs every Claude Code tool call through a local rule and decides. ```bash npx @ordo-engine/cli guard init # scaffold the policy + wire the Claude Code hook ``` Now your agent is gated. When it reaches for something destructive, Ordo stops it — with a reason: ```text $ (agent tries) rm -rf ./build ⛔ Denied by policy: Destructive shell command blocked by policy [policy@1.0.0 · DENY] ``` The policy is a **normal Ordo project** in `.ordo-guard/` — so your guardrails have a test suite: ```bash ordo guard test # ✔ blocks rm -rf ✔ asks before git push ✔ allows read-only git … ordo guard log # every decision, timestamped and auditable ``` Edit `.ordo-guard/rulesets/policy.json` in plain expressions (`tool == 'Bash' && command contains 'terraform destroy'`), add a test, ship. → **[Guard docs](https://ordo-engine.github.io/Ordo/docs/en/platform/guard)** --- ## The same engine, everywhere decisions happen Guardrails are one use of one engine. The same rules — sub-microsecond, JIT-compiled — run as an embedded call, an HTTP/gRPC service, or in the browser via WASM, and a full **platform + visual Studio** lets business teams author and govern them without touching code. **Engine** — sub-microsecond rule execution, JIT-compiled (Cranelift), runs everywhere (HTTP · gRPC · WASM · CLI). **Studio** — visual flow editor, decision-table editor, trace-driven testing, version history. **Platform** — org workspace, project hub, server fleet, release center, testing and governance. <p align="center"> <img src="images/screenshots/first_page.png" alt="Ordo Platform Dashboard" width="100%" /> </p> --- ## Why Ordo? | | **Ordo** | OPA | Drools | json-rules-engine | |---|---|---|---|---| | JIT compilation | ✅ Cranelift | ❌ | ❌ | ❌ | | Testable agent guardrails | ✅ `ordo guard` | ⚠️ policy-only | ❌ | ❌ | | Authoring model | Visual flow + decision table | Rego policy code | DRL / DMN + Business Central | JSON rule DSL | | Built-in web workbench | ✅ Studio | Playground / APIs | ✅ Business Central | ❌ | | Browser / Wasm target | ✅ Native | ✅ Policy-to-Wasm | ❌ | ✅ Browser JS | | Deployment | Single binary or hosted platform | Binary, sidecar, or service | JVM app, KIE Server, Business Central | JS library for Node/browser | Comparison focuses on first-party documented authoring and deployment capabilities rather than cross-project benchmark claims. --- ## More ways to run it ### Author rules as files (local, offline) ```bash npx @ordo-engine/cli init my-rules && cd my-rules ordo validate && ordo test # sub-second, no network ordo trace loan-approval --input '{"amount":5000}' ``` ### Try Studio (visual, 5 minutes) ```bash docker compose up # platform + engine + studio # open http://localhost:5173 ``` Or use the hosted **[Live Playground](https://ordo-engine.github.io/Ordo/)** — no install needed. ### Engine as a service ```bash docker run -d -p 8080:8080 ghcr.io/pama-lee/ordo:latest # Create a rule curl -X POST http://localhost:8080/api/v1/rulesets \ -H "Content-Type: application/json" \ -d @examples/discount.json # Execute curl -X POST http://localhost:8080/api/v1/rulesets/discount/execute \ -d '{"input": {"user": {"vip": true}}}' # → {"code": "VIP", "message": "20% discount"} ``` ### Embed the editor ```bash npm install @ordo-engine/editor-vue # Vue 3 npm install @ordo-engine/editor-react # React ``` ### Call from your app ```bash go get github.com/pama-lee/ordo-go # Go pip install ordo-sdk # Python # Java (Maven): com.ordoengine:ordo-sdk-java ``` --- ## Platform Features ### Rule Authoring - **Flow editor** — drag-and-drop decision trees with live WASM execution - **Decision table** — spreadsheet-style multi-condition rules - **Templates** — ship pre-built rule sets (e.g. `ecommerce-coupon`) that teams clone in one click ### Platform Workspace - **Organization and project hub** — manage orgs, projects, members, roles, and project workspaces from one dashboard - **Server fleet** — register connected Ordo servers, group them by environment, and target release rollouts by instance - **Release center** — create release requests, track staged rollouts, approvals, retries, and rollback history ### Governance - **Fact catalog** — register every input field with type, source, latency, and null policy - **Decision contracts** — typed input/output schemas per ruleset; generate test skeletons automatically - **Version history** — full snapshot log per ruleset; roll back with one click - **Test and diagnostics** — run platform-local draft tests, inspect expected/actual diffs, replay execution paths, and review decision-table hits ### Engine - **1.63 µs** interpreter · **50–80 ns** JIT (Cranelift) · **54k QPS** HTTP single-thread - Multi-tenancy, WAL crash recovery, hot reload, NATS JetStream sync - Data Filter API — push rule logic into SQL / MongoDB `$match` / JSON predicates - Compiled `.ordo` binary format with ED25519 signature for rule protection --- ## Project Structure ``` ordo/ ├── crates/ │ ├── ordo-core/ # Rule engine + JIT compiler │ ├── ordo-server/ # HTTP/gRPC API server │ ├── ordo-platform/ # Org/project/template/testing management layer │ ├── ordo-cli/ # CLI — guard, init, validate, test, trace, mcp │ ├── ordo-wasm/ # WebAssembly bindings │ ├── ordo-proto/ # gRPC definitions │ └── ordo-derive/ # TypedContext derive macro ├── ordo-editor/ │ ├── packages/ # @ordo-engine/editor-{core,vue,react,wasm} │ └── apps/ │ ├── studio/ # Platform Studio (Vue 3 + TDesign) │ ├── playground/ # Live demo │ └── docs/ # VitePress documentation ├── sdk/ # Official Go / Python / Java client SDKs └── scripts/ # sync-version.sh, benchmarks ``` --- ## License MIT — see [LICENSE](LICENSE). <p align="center">_{Built with Rust · <a href="https://discord.gg/Y529FkArhh">Discord</a> · <a href="https://ordo-engine.github.io/Ordo/">Docs</a>}</p>