AI orchestration, edge-native. Built exclusively for Cloudflare.

Build AI workflows that you actually control. No proprietary platforms, no vendor lock-in, no serverless sprawl. Just Git, YAML, and Cloudflare Workers—with zero cold starts across 300+ global locations.

Not cloud-agnostic. Purpose-built for Cloudflare's edge infrastructure: Workers, KV, D1, R2, Durable Objects, Queues, Vectorize, and Workers AI.

You have 50+ prompts, 20+ agents, and dozens of SQL queries scattered across your codebase. Changing one breaks three others. Rollbacks mean digging through Git history. A/B testing means deployment hell. And every "AI orchestration platform" wants you to give up control for their black box.

You know there's a better way.

Version every AI component independently. Deploy any combination from any point in history.

# Each component versions independently
edgit tag create extraction-prompt v1.0.0
edgit tag create company-agent v2.1.0
edgit tag create validation-sql v0.5.0
edgit tag create invoice-schema v1.0.0

# Deploy optimal versions from different timelines
edgit tag set extraction-prompt prod v0.1.0
edgit tag set company-agent prod v3.0.0
edgit tag set invoice-schema prod v1.0.0
edgit push --tags --force

# Instant rollback
edgit tag set extraction-prompt prod v0.1.0
edgit push --tags --force  # < 50ms globally

No merge conflicts. No monorepo versioning. No JSON state files.

Edgit creates and manages git tags. That's it. GitHub Actions handles deployment. All version data lives in Git tags. Your CI/CD deploys to Cloudflare KV. Every version ever created is instantly accessible at the edge.

→ Get Started with Edgit

Compose AI workflows in YAML. Execute on Cloudflare's global network—300+ locations, zero cold starts, V8 isolates not containers.

# ensembles/company-intelligence.yaml
name: company-intelligence

trigger:
  - type: http
    path: /api/company-intel
    methods: [POST]
    public: true

flow:
  - name: fetch-company-data
    agent: fetch-company-data
    input:
      domain: ${input.domain}
    cache:
      ttl: 3600

  - parallel:
    - name: analyze-financials
      agent: analyze-financials

    - name: fetch-competitors
      agent: fetch-competitors

output:
  data: ${fetch-company-data.output}
  analysis: ${analyze-financials.output}

No DAG builders. No UI-driven workflows. No central orchestrator bottleneck.

Define workflows in Git. Deploy to Cloudflare Workers. State management, caching, and parallelization built in.

→ Learn About Conductor

# Version components with Edgit
components:
  extraction-prompt: v1.0.0
  company-analyzer: v2.1.0

# Load as Conductor agents
agents/
  fetch-company-data.yaml
  analyze-financials.yaml
  fetch-competitors.yaml

# Orchestrate in ensembles
ensembles/
  company-intelligence.yaml

Mix optimal component versions. Test combinations locally. Deploy atomically to the edge.

Managed service with generous free tier for:

Prompt & Component Management

• Visual editor for prompts, SQL, configs
• Version and tag changes with one click
• A/B test and multivariant experiments
• Let analysts iterate without touching code

Workflow & Deployment

• Visual workflow builder (React Flow)
• Deploy to production from the UI
• Real-time collaboration across teams
• Observability dashboards

The Difference: Your data lives in Git, not our database. We're the UI layer. You own the source of truth.

Think PromptLayer's editing experience, but Git-native instead of proprietary storage.

Cloudflare Workers, KV, D1, R2, and AI Gateway are the primitives. No centralized compute.

Configuration, orchestration, and versioning live in Git. CLI and SDK are thin layers around Git operations.

Multi-layer caching with TTL control. Agent-level cache settings. Cache-first thinking reduces costs and latency.

LLM and API agents produce machine-readable, type-safe output validated via JSON schema.

Every execution emits structured logs and metrics. Debugging and transparency are effortless.

Core tooling (Edgit, Conductor) is open source. Cloud is proprietary—we charge for the UI and managed service, not the runtime.

Edgit (active development)

• ✅ Component versioning (prompts, queries, configs, scripts, schemas)
• ✅ Deployment management
• ✅ AI-powered commits
• ✅ Smart detection
• ✅ Component listing (tree, JSON, YAML, table)
• ✅ Discovery tools (scan, detect, patterns)

Conductor (production-ready)

• ✅ Core runtime with graph executor
• ✅ State management (immutable, access tracking)
• ✅ Triggers: http, webhook, mcp, email, queue, cron, build, cli (8 types)
• ✅ Operations: think, code, storage, data, http, tools, email, sms, html, pdf, form, queue, docs (13 types)
• ✅ Structured outputs with JSON Schema components
• ✅ Durable Objects (ExecutionState, HITL)
• ✅ Webhooks (inbound/outbound with HMAC signatures)
• ✅ Scheduled execution (cron triggers)
• ✅ Comprehensive test suite (1500+ tests passing)
• ✅ CLI tools (exec, agents, test, docs, history, logs, state, replay, health, config)
• ✅ SDK with client library & testing utilities
• ✅ Observability & logging
• ✅ AI Gateway integration
• ✅ Scoring system for quality metrics

Cloud (design phase - managed service with generous free tier)

# Launch the wizard (no installation needed)
npx @ensemble-edge/ensemble

# Or create a Conductor project directly
npx @ensemble-edge/ensemble conductor init my-project
cd my-project

# Install and start dev server
pnpm install
ensemble conductor start

The Ensemble CLI provides access to all tools: Conductor (orchestration), Edgit (versioning), and Cloud (managed platform).

# Initialize Edgit in your project
npx @ensemble-edge/edgit init

# Version a prompt
npx @ensemble-edge/edgit tag create my-prompt v1.0.0

# Deploy to staging
npx @ensemble-edge/edgit tag set my-prompt staging v1.0.0
npx @ensemble-edge/edgit push --tags --force

# Deploy to Cloudflare Workers
npx wrangler deploy

# Execute ensemble
curl https://your-worker.workers.dev/api/hello \
  -H "Content-Type: application/json" \
  -d '{"name": "World"}'

For automated environments, use the -y flag to skip interactive prompts:

npx @ensemble-edge/conductor init my-project -y
npx @ensemble-edge/edgit init -y

Modern AI development looks like this:

1. Ship v1.0.0 of your app
2. Change a prompt
3. Everything becomes v2.0.0
4. Original prompt trapped in Git history
5. Can't A/B test old vs new
6. Can't mix optimal versions
7. Rollback means reverting the entire codebase

This is insane for systems with 100+ independently evolving components.

Ensemble Edge treats AI components like they deserve: individually versioned, independently deployable, infinitely composable.

Then orchestrates them at the edge where they execute fast and scale infinitely.

📚 Complete Documentation - Comprehensive guides and tutorials

• Introduction - Core concepts and overview
• Edgit - Component versioning system
  • Installation
  • Versioning Guide
  • CLI Reference
• Conductor - Edge orchestration framework
  • Your First Project
  • Operations
  • Starter Kit
  • Playbooks
• Cloud - Managed service (coming soon)
• Examples - Real-world implementations

We're building in public. Watch development at:

• Edgit Repository - Component versioning
• Conductor Repository - Edge orchestration
• Documentation - Mintlify docs site

Issues, PRs, and feedback welcome.

• Edgit: MIT
• Conductor: Apache 2.0
• Documentation & Examples: MIT
• Ensemble Cloud: Proprietary (managed service)

Open core, closed cloud. Use it, fork it, build on it.

Built by engineers who believe AI tooling should be as solid as the infrastructure it runs on.

No buzzwords. No hand-holding. Just tools that work.