DAG Orchestration
Declarative JSON-based directed acyclic graphs with parallel lanes, hard barriers, and supervisor checkpoints — zero boilerplate.
Learn more →✅ Production-Ready424 tests passingMIT
Automate AI workflows
without the complexity
Enterprise-grade multi-agent orchestration engine — DAG-supervised parallel agents with streaming LLM output, intelligent model routing, resilience patterns, RBAC, audit logging, and a zero-API-key demo mode.
Terminal
npm install -g @ai-agencee/ai-kit-cli
# zero-API-key demo — see the engine run in < 30 s
git clone https://github.com/binaryjack/ai-starter-kit
cd ai-starter-kit && pnpm install && pnpm demo424Tests passing
13Enterprise features (E1–E13)
7LLM providers
$0Cost to evaluate (mock mode)
Capabilities
Everything for enterprise AI workflows
12 production-grade capabilities — orchestration, resilience, compliance, and developer tooling — all in a single zero-configuration package.
Orchestration & Execution
Intelligent Model Routing
Automatically selects the optimal model tier (Haiku → Sonnet → Opus) based on task complexity and remaining budget.
Learn more →Resilience Patterns
Exponential-backoff retry, per-provider circuit breakers, and graceful fallbacks keep workflows running through transient failures.
Learn more →Real-Time Streaming
Token-by-token LLM output streamed live to stdout — every provider, including the built-in Mock provider.
Learn more →Enterprise & Security
RBAC & OIDC Auth
Role-based access control with RS256/ES256 JWT validation. Every run is principal-tagged and GDPR-ready.
Learn more →Immutable Audit Logging
Hash-chained tamper-proof audit logs for every agent action — compliance-ready out of the box.
Learn more →Multi-Tenant Isolation
Hard filesystem and runtime isolation per tenant — each run lives in its own scoped directory tree.
Learn more →PII Scrubbing
Automatic regex-based detection and redaction of emails, phone numbers, SSNs, and API keys before they reach LLM providers.
Learn more →Developer Experience
Powerful CLI
Full-featured command-line tool — init, sync, check, run DAGs, plan entire projects, visualise graphs.
Learn more →MCP Integration
Native Model Context Protocol bridge — connect ai-agencee directly to Claude Desktop or VS Code Copilot.
Learn more →TypeScript Builder API
Fluent, fully type-safe DSL for constructing DAGs in code — no JSON required.
Learn more →Observability
Typed Event Bus
Subscribe to real-time DAG lifecycle events — token streams, cost updates, lane status — with full TypeScript typing.
Learn more →Plan System
From vague idea to running code
The 5-phase plan system takes a raw requirement through BA-led discovery, parallel decomposition, dependency wiring, and DAG execution — with every agent knowing their scope before writing a line.
- 0
Phase 0 — DISCOVER
BA agent interviews you with ~12 structured questions across problem definition, primary users, stories, and stack constraints.
Input: Your idea (free text)Output: discovery.json — complete DiscoveryResultbusiness-analyst - 1
Phase 1 — SYNTHESIZE
BA reads the discovery result and produces a plan skeleton — Steps with rough Tasks, ownership, and acceptance criteria.
Input: discovery.jsonOutput: plan.json (phase: synthesize)business-analyst - 2
Phase 2 — DECOMPOSE
Each specialist agent expands their Steps into fully-detailed Tasks in parallel — description, acceptance criteria, effort, artefacts.
Input: plan.json (synthesize)Output: plan.json (fully populated)architecturebackendfrontendtestinge2e - 3
Phase 3 — WIRE
Dependencies between tasks are resolved — shared contracts agreed, API schemas locked, integration points mapped.
Input: plan.json (decompose)Output: plan.json (wired)architecturebackendfrontend - 4
Phase 4 — EXECUTE
The wired plan is converted to a DAG and handed off to the orchestration engine for parallel execution.
Input: plan.json (wired)Output: DAG run results + code artefactsbackendfrontendtestinge2esupervisor
Agent Roster
Specialised agents, coordinated by the DAG
Each agent is a focused expert. The DAG engine assigns tasks, enforces quality checkpoints, and routes failures — so agents never need to coordinate manually.
Business Analyst
Discovery & requirementsDrives the 5-phase discovery process — interviews you with structured questions and synthesises a precise sprint plan before a single line of code is written.
Requirements analysisFeature breakdownAcceptance criteriaEffort estimationStakeholder translation
Architect
System & schema designDesigns the system architecture, data schemas, and API contracts — producing ADRs (Architecture Decision Records) at Opus-tier quality.
System designData schemaAPI contractsADR generationDependency analysis
Backend
API & infrastructureImplements server-side code, API handlers, database migrations, and integration tests — guided by the architect's ADRs.
API implementationDatabase migrationsAuth integrationUnit test generationPerformance profiling
Frontend
UI & component designBuilds React components, wires state, implements design tokens — running in parallel with the Backend agent behind a soft-align barrier.
React component authoringState managementDesign token usageAccessibilityResponsive layout
Testing
Unit & integration testsGenerates comprehensive unit and integration tests, asserting acceptance criteria from the sprint plan.
Unit test generationIntegration test stubsCoverage analysisMock generationTest plan authoring
E2E
End-to-end validationValidates the complete user flow from browser to database — blocked on Backend + Frontend completion via a hard barrier.
Playwright / Cypress scenariosCritical path coverageRegression suitesCI gate authoring
Supervisor
Quality checkpointValidates lane output deterministically at every checkpoint — issues PASS, RETRY (with injected guidance), HANDOFF, or ESCALATE verdicts.
Deterministic checksLLM review checksRetry injectionHuman escalationBudget gating
Model Routing
Right model, right cost
The router automatically selects the cheapest model tier that satisfies each task's complexity requirement — and falls back to lower tiers when the budget is running low.
| Task type | Tier | Anthropic | OpenAI | Cost / 1 M tokens |
|---|---|---|---|---|
| file-analysis | haiku | claude-haiku-4-5 | gpt-4o-mini | $0.80 |
| code-generation | sonnet | claude-sonnet-4-5 | gpt-4o | $3.00 |
| code-review | sonnet | claude-sonnet-4-5 | gpt-4o | $3.00 |
| architecture-decision | opus | claude-opus-4-5 | gpt-4o | $15.00 |
| security-review | opus | claude-opus-4-5 | gpt-4o | $15.00 |
Mock provider has zero cost and requires no API key — use it for evaluation, CI, and testing. Custom providers (Ollama, Bedrock, Gemini) available on Enterprise.
Why ai-agencee
Built for production workflows,
not chat prompts
| Capability | Generic AI chat | Code-gen copilots | ai-agencee |
|---|---|---|---|
| Structured multi-step plan from a vague idea(5-phase BA-led discovery → wired sprint plan) | ✗ | ⚠ | ✓ |
| Parallel agent coordination with sync points(DAG barriers, soft-align, read-contract) | ✗ | ✗ | ✓ |
| Automatic retry + escalation on failure(retryBudget, HANDOFF, ESCALATE verdicts) | ✗ | ✗ | ✓ |
| Human-in-the-loop approval gates(needs-human-review checkpoint) | ✗ | ✗ | ✓ |
| Enterprise: RBAC, audit, multi-tenant, PII, OIDC(E1–E13 enforced at runtime) | ✗ | ✗ | ✓ |
| Zero-cost evaluation + CI integration(Mock provider, $0.00, no keys) | ✗ | ✗ | ✓ |
| Extensible: custom agents, checks, providers(Plugin system + TypeScript Builder API) | ⚠ | ⚠ | ✓ |
| Per-run cost tracking & budget enforcement | ✗ | ✗ | ✓ |
| Real-time streaming output(Every provider, including Mock) | ⚠ | ⚠ | ✓ |
| MCP / Claude Desktop integration(Native MCP server, zero extra config) | ✗ | ⚠ | ✓ |
Get Started
Up and running in under 5 minutes
The CLI, mock provider, and DAG engine are open source and free forever. No sign-up, no API key, no billing details needed to evaluate.
1. Install CLI
npm install -g @ai-agencee/ai-kit-cliOr: pnpm add -g @ai-agencee/ai-kit-cli
2. Run the zero-key demo
git clone https://github.com/binaryjack/ai-starter-kit
cd ai-starter-kit
pnpm install && pnpm demoMock provider — no API keys, no cost
3. Run a real DAG
ANTHROPIC_API_KEY=sk-... ai-kit agent:dag ./my-dag.jsonOr use OpenAI: --provider openai
4. Start a planning session
ai-kit plan5-phase BA-led discovery → sprint plan → DAG
Use it programmatically
Drop the engine directly into any Node.js / TypeScript project.
npm install @ai-agencee/ai-kit-agent-executorAPI docs →Ready to start?
Build your first multi-agent workflow
in under 5 minutes
No API key, no credit card. Clone the repo and run pnpm demo to see DAG-supervised agents in action.