// WHY HARPER

Don’t just prototype. Build production software.

Agents can generate code in seconds. Getting that code to production is the hard part, it needs infrastructure, security, and operational trust. Harper gives agents a contained runtime where the entire stack is in the code. What they build locally is what you deploy. No credentials to hand over, no services to wire up, no gap between prototype and production.

Rocket icon launching beside a glowing cube, representing performance and acceleration.

Ship Fast Apps Fast

Harper gets you to production faster, and what you deploy already performs at scale. No gap between prototype speed and production performance because it's the same runtime the whole way through. No glue code, service orchestration, database wiring, or manual infra. Harper collapses all of that into one runtime.

Go to Repo

Network node icon connected to a glowing cube, representing distributed architecture.

Scale Without Overhead

Deploying AI-generated applications across regions shouldn't require a platform team. Harper Fabric deploys your application across distributed geographies in minutes, increasing capacity, throughput, and resilience while driving down global latency as new regions come online.

Shield icon over a glowing cube, representing enterprise-grade security and protection.

Enterprise Proven

Most enterprises can’t hand agents the keys to their cloud provider, database, or caching layer. With Harper, they don’t have to. The entire stack runs locally in a single runtime, agents never touch production infrastructure directly. Trusted by some of the world’s largest enterprises.

// GET STARTED

Agentic Engineering with Harper

1st: Scaffold ↓

2nd: Launch Agent ↓

Terminal

$ cd my-app

✓ navigate to your project
‍

$

npm create harper@latest
‍

✓ harper installed
‍

‍✓ skills added

Claude Code

$ claude

Claude: reads skills/harper-best-practices on startup.

‍> Add vector search to the product table with semantic similarity on descriptions

✓ Updated schema.graphql | description: [Float] @indexed(type: "HNSW")
✓ Created resources/search.ts using Harper's native vector search API
✓ No external vector DB. Runs in-process

OpenAI Codex

$ codex

Codex: indexes the full project including skills/ for Harper-specific patterns.

‍> Set up a caching layer that pulls inventory data from our existing REST API

✓ Updated schema.graphql | type Inventory @table @cache(expiration: 3600)
✓ Created resources/inventory.ts get() fetches from external API, Harper caches
✓ ETag + stampede prevention built in

Google Antigravity

$ antigravity

Antigravity: loads skills/ alongside your schema and resource definitions.

‍> Add real-time order notifications over WebSocket when orders are created

✓ Updated resources/orders.ts publish('orders', record) on POST
✓ Created client/useOrders.ts subscribes via Harper's native pub/sub
✓ No message broker — WebSocket built in

Use Your Own AI Agent

Already using Claude Code, Codex, or Antigravity? Skills files give your existing agent full Harper runtime knowledge. No switching required.

Scaffold

npm create harper@latest adds Harper and its best practices to your project. Works with any agent that reads project-level files for context.

Launch

Claude Code, OpenAI Codex, and Google Antigravity all read project-level files as context. The skills/ directory is picked up automatically. No setup needed.
Works with any agent that reads project files for context.

Build

Your agent knows the exact syntax for vector search, real-time pub/sub, caching, auth, and deployment. No multi-service architecture, no drift. Works with your existing CI/CD pipeline.

View Repo on GitHub

Try Harper Agent

OPEN ALPHA

Where skills teach external agents about Harper, Harper Agent is built to think in it. It scaffolds and codes entire projects contextually, generating structurally sound applications grounded in Harper's architecture from the first command.

Bring your own model API key to use this CLI-based tool. Harper Agent supports OpenAI, Anthropic, Google Gemini, and local models via Ollama.

Generates Harper schemas, resources, and config grounded in runtime conventions

Detects existing Harper projects and works within established structure

Supports session persistence for long-running development workflows

Non-interactive mode for piping prompts from specs or task files

Install Harper Agent

Terminal

# Install globally
$ npm install -g @harperfast/agent

# Start in your project directory
$ harper-agent

Working directory: ~/Code/my-app
Harper app detected: Yes

Harper: What do you want to do together today?

> Create a support knowledge base that stores documents as blobs, indexes them with vector embeddings for semantic search, and pushes new article alerts via pub/sub.

# Or pipe a spec non-interactively
$ cat spec.md | harper-agent

Built Different

For most stacks, infrastructure lives outside the code, agents can’t reach it safely. On Harper, the entire application surface is declarative and in-repo so agents control everything without touching production infra.

Layered Harper application platform stack showing GraphQL, API, REST, WebSocket, and MQTT on top of cache and in-memory layers, with database, NoSQL, blob, and vector storage beneath, and AI coding tools above.

TYPICAL
ARCHITECTURE

Configuration Model

config.yamlDeclrative, editable by AI

Distributed configs (Terraform, cloud consoles, service dashboards)

API Definition

schema.graphqlAuto-generated API layer

Custom server setup + framework wiring

Business Logic

resource.jsExecuted within unified runtime

Separate service, separate runtime, separate deploy

Database Provisioning

Built-in (NoSQL, Blob, Vector, Cache)

Manual provisioning (MongoDB, Postgres, Redis, etc.)

Deployment Model

Harper Fabric deploys directly from Harper configuration

CI/CD pipelines + cloud infra coordination

Multi-Region Deployment

Native, built into Harper Fabric

Complex setup across providers

AI Agent Control Surface

Entire stack accessible through files AI can modify

Infra lives outside code; AI cannot provision safely

Human-in-the-Loop Required

Yes (DevOps + cloud permissions required)

Operational Complexity

Minimal, single runtime

High, service coordination and integration overhead

P95 Server Latency

1–10 ms

50–200 ms typical

Production Readiness

Includes all core services

Requires stitching multiple services together

Open Source

Yes

Varies

What Makes This Possible?
Declarative files give agents full control of the stack, without full access to your credentials or data.

Unified Runtime for Agentic Engineering