Skip to main content
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. You’re an engineer. You deserve tools that treat you like one.

​
The Problem

You’ve got 50+ prompts, 20+ agents, dozens of SQL queries scattered across your codebase. Change one, break three others. Rollback means Git archaeology. A/B testing means deployment hell. And every “AI orchestration platform” wants you to hand over control to some black box that the head of sales bought because the pitch deck looked good. You know there’s a better way.

​
Components Agents Ensembles

The building blocks are simple:

Components

Versioned artifacts: prompts, configs, queries, scripts. Each gets its own version history. Mix and match optimal versions from different points in time.

Agents

Workers that execute: AI reasoning, database queries, HTTP calls, business logic. Each agent uses operations and components to get work done.

Ensembles

Orchestration workflows: YAML files that coordinate agents into complex workflows. Execute at the edge in <50ms cold starts.

​
Edgit: Git-Native Component Versioning

Version every AI component independently. Deploy any combination from any point in history. Rollback in <50ms globally. 
# Each component versions independently
edgit tag create extraction-prompt v1.0.0
edgit tag create company-agent v2.1.0
edgit tag create validation-query v0.5.0

# Deploy the optimal combination - cherry-pick from history
edgit tag set extraction-prompt prod v0.1.0  # Ancient but perfect
edgit push --tags --force

edgit tag set company-agent prod v3.0.0      # Latest stable
edgit push --tags --force

edgit tag set validation-query prod v2.5.0   # Best performance
edgit push --tags --force

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

Start versioning components in 5 minutes

​
Conductor: YAML-Driven Edge Orchestration

Define AI workflows in YAML. Execute at 300+ global locations on Cloudflare Workers. Zero cold starts via V8 isolates. No DAG builders, no UI-driven workflows, no central orchestrator bottleneck. 
# ensembles/company-intelligence.yaml
name: company-intelligence

agents:
  - name: fetch-company-data
    operation: http
    config:
      url: https://api.example.com/companies/${input.domain}
      cache_ttl: 3600

  - name: analyze-financials
    operation: think
    component: [email protected]
    config:
      model: claude-3-5-sonnet-20241022
    input:
      company_data: ${fetch-company-data.output}

  - name: generate-report
    operation: think
    component: [email protected]
    input:
      analysis: ${analyze-financials.output}
Define workflows in Git. Deploy to Cloudflare Workers. State management, caching, parallelization built-in.

Learn About Conductor

Edge orchestration for AI workflows

​
The Versioning Multiverse

With independent versioning of components AND agents, you unlock infinite combinations: 
# Same ensemble, different realities running in parallel
ensemble: analysis-pipeline

agents:
  # Production: stable + conservative
  - name: prod-analyzer
    agent: [email protected]        # Stable agent version
    component: [email protected]      # Conservative prompt

  # Experimental: new everything
  - name: exp-analyzer
    agent: [email protected]   # New agent logic
    component: [email protected]      # Aggressive prompt

  # Hybrid: stable agent, new prompt
  - name: hybrid-analyzer
    agent: [email protected]        # Stable agent
    component: [email protected]      # New prompt only
Test: 2 agent versions 3 prompt versions 2 configs = 12 variants running simultaneously. Perfect for A/B testing, progressive rollouts, and reproducing bugs from last Tuesday at 2:47 PM.

​
Why This Exists

Modern AI development is broken:
  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.

​
Architecture Principles

Purpose-built for Cloudflare’s edge infrastructure: Workers, Workers AI, KV, D1, R2, Durable Objects, Queues, and Vectorize. No centralized compute. No single point of failure.
Configuration and orchestration and versioning live in Git. No proprietary storage. You own the source of truth.
Multi-layer caching with TTL control. Agent-level cache settings. Cache-first thinking reduces costs and latency by 10x.
AI operations produce machine-readable and type-safe output validated via JSON schema. No string parsing and no hallucinated JSON.
Every execution emits structured logs and metrics. Debugging and monitoring are effortless and not afterthoughts.
Core tooling (Edgit and Conductor) is open source. We charge for managed services and not lock-in.

​
Quick Start

Get Started

npx @ensemble-edge/ensemble
# Or: npx @ensemble-edge/ensemble conductor init my-project

Version Components

npx @ensemble-edge/edgit tag create my-prompt v1.0.0
npx @ensemble-edge/edgit tag set my-prompt prod v1.0.0
npx @ensemble-edge/edgit push --tags --force

Deploy to Edge

npm run build
npx wrangler deploy
# Live at https://your-worker.workers.dev

​
Status

Edgit

Active developmentComponent & agent versioning Deployment management Git-native tags Discovery tools

Conductor

Production-readyEdge orchestration 10+ operations 8+ pre-built agents State management A/B testing 276 tests passing

Cloud

Design phaseManaged service with generous free tier (coming soon)

​
Next Steps

Core Concepts

Understand the building blocks

5-Minute Quick Start

Get running in 5 minutes

Edgit Docs

Component versioning and deployment

Conductor Docs

Edge orchestration framework

Playbooks

Real-world patterns and recipes

Community

GitHub discussions and support
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.