⚙️ Forge v2.1 — Now an AI Operating System for autonomous software engineering · Multi-agent orchestration · MCP-native · Defense-grade privacy. Read the platform brief →
v2.1 · The AI Operating System · 36 modules · MCP-native

The AI Operating System
for Autonomous Software Engineering

Forge is not another AI code editor. It is a complete development platform — multi-agent orchestration, MCP-native extensibility, defense-grade privacy, and a browser-native runtime that runs your code, your agents, and your deployments with zero servers and zero tracking.

7+
AI Providers
36
Modules
AES-256
Encryption
$0
Server Costs
100%
Client-Side
forge.html — Multi-Agent Workspace
1// Spawn 3 agents with different models, working in parallel
2const architect = new ForgeAgent({ name: 'Architect', model: 'claude-opus-4-8' });
3const frontend = new ForgeAgent({ name: 'Frontend', model: 'gpt-5.5' });
4const backend = new ForgeAgent({ name: 'Backend', model: 'deepseek-v4' });
5
6// They share state, lock files, and build together — securely
7await architect.run('Design a REST API for a todo app');
8await frontend.run('Build the React UI components');
9await backend.run('Implement the Node.js endpoints');
10
11// Deploy to Netlify in one click — self-healing checks first
12await ProjectManager.deployNetlify(); // ✅ Live in seconds

Not an editor.
An AI Operating System.

A code editor edits files. Forge orchestrates an entire software-engineering lifecycle — agents, tools, data, deployments, and plugins — under one browser-native runtime with a layered architecture inspired by operating-system design.

L7 · Application

Application Layer

What the user touches: visual workspaces, no-code builders, template mode, real-time multiplayer, extension marketplace, deploy pipeline. The same runtime that ships a landing page also orchestrates a 25-iteration autonomous coding loop — without the user switching tools.

MultiplayerNo-Code BuildersTemplate ModeMarketplace1-Click Deploy
L4 · Orchestration

Orchestration Layer — the core differentiator

Where most tools give you one agent in one chat, Forge runs a fleet. Spawn any number of agents on different models, give them roles, and they coordinate via shared state, file locks, and a real-time execution graph. Token budgets and rate limits are enforced here — not in your wallet after the fact.

Multi-AgentExecution GraphFile LockingShared StateToken BudgetAgent Evals
L3 · Runtime

Runtime Layer

A full browser-native execution environment. WebContainer runs Node.js, Python, and shell commands without leaving the page. WASM isolates untrusted plugins. SQLite and a KV store persist structured agent state. Deploy targets (Netlify, GitHub) are first-class citizens, not afterthoughts.

WebContainerWASM SandboxSQLite + KVOllama / WebGPUNetlify / GitHub
L2 · Security

Defense-Grade Security Layer

API keys never leave the device. Cross-device sync is end-to-end encrypted (AES-256-GCM, PBKDF2 100K iterations). Plugins run in sandboxed iframes with no same-origin access. Zero telemetry, zero analytics, zero accounts. Built for environments where privacy is a hard requirement, not a marketing line.

AES-256-GCMSandbox iframeDOMPurifyZero TelemetryAir-Gappable
L1 · Kernel

Forge Kernel

The core that ties it together: a unified provider interface (Claude, GPT, Gemini, DeepSeek, Groq, Ollama, WebGPU), the MCP client + server, the plugin host, the project manager, and the on-device key vault. The kernel is the reason Forge can run fully offline, fully air-gapped, or fully cloud — your choice.

7 ProvidersMCP HostPlugin HostProject ManagerKey Vault

Everything you need to build
with AI — in one workspace

Forge combines the best of Claude Code, Cursor, and Codex into a single, secure, browser-based workspace. No terminal required. No data leaves your device.

Multi-Agent Collaboration

Spawn multiple AI agents with different models. They lock files, share state, and build together in parallel — without stepping on each other's toes.

Unique

Sandboxed Plugins (WASM)

Custom plugins run in isolated iframes — they can never touch your API keys, localStorage, or cookies. WebContainer-grade security, zero friction.

v2.1

Cross-Device Sync

Start on your laptop, continue on your phone. End-to-end encrypted with AES-256-GCM. API keys never leave your device.

AI Code Completion

Copilot-style inline completions. Word-based + AI-powered. Uses your API key — no middleman, no subscription.

Extension Marketplace

Browse, install, and submit community extensions. One-click install with ratings, categories, and source code review.

Offline-First Models

Run Llama 3, Mistral, or Phi via Ollama or WebGPU when internet is spotty. Auto-fallback when cloud APIs fail.

v2.0

Self-Healing Agent

Background linter that detects and auto-fixes syntax errors before you push. No more failed builds from typos.

Shared Agent State

SQLite + KV store in the browser. Agents persist structured data across sessions without bloating chat context.

Execution Graph

Visualize exactly which agent called what tool and when. Debug complex multi-agent loops with a node map.

MCP Server & Client

Expose your workspace as an MCP server. Connect Notion, Google Drive, GitHub, Slack — with one-click OAuth.

v2.1

Token Budget

Set hard USD budgets and rate limits. Visual cost tracking per model, per agent. Never get a surprise API bill again.

Real-Time Multiplayer

Code together with live cursors, presence avatars, and shared agent activity. Like Figma for AI agents.

v2.0

Autonomous Coding

Agents plan, implement, verify, and document complex multi-step tasks autonomously. 25-iteration loops. Like Claude Code — but visual.

v2.1

Agent Evals Suite

Benchmark your agents with standardized test cases. Track pass rates, latency, and cost across model changes.

No-Code Builders

Non-devs create skills, plugins, and agents with plain English. Visual form builders. Zero code required.

v2.1

Template Mode Preview

Build landing pages, cards, and forms with side-panel customization. No code — just colors, text, and layout toggles.

v2.1

Multi-Agent Orchestration,
not just multi-chat

Other tools let you talk to one model. Forge lets you run a coordinated team of them — each with a role, a model, and a budget — all working on the same project at the same time. This is what makes Forge an operating system, not an editor.

You — single workspace, single source of truth
Architect
claude-opus-4
Frontend
gpt-5.5
Backend
deepseek-v4
File Locksarchitect → /api/*
frontend → /ui/*
Shared StateSQLite · KV store
across all agents
Execution Graphevery tool call
logged & visualized
Token Budget$2.50 / iteration
hard limit enforced
  • 1
    Role-based agents, not generic chats

    Spawn an Architect on Claude for reasoning, a Frontend on GPT-5.5 for speed, a Backend on DeepSeek for cost. Each agent gets a name, a system prompt, a model, and a file scope. They are not three chat tabs — they are three coordinated workers.

  • 2
    File locking prevents collisions

    When the Architect claims /api/*, the Frontend is blocked from writing there. No more two agents clobbering each other's edits. Locks are visual, revocable, and auditable in the execution graph.

  • 3
    Shared state via SQLite + KV

    Agents persist structured data — schemas, TODO lists, decisions, intermediate results — without bloating their chat context. The Architect can leave a note for the Backend in a shared table; the Backend reads it on its next iteration. This is how a real engineering team works.

  • 4
    Hard token budgets, not regrets

    Set a USD ceiling per run, per agent, or per project. The orchestrator force-stops the loop the moment the budget is hit. You see cost per model, per agent, per request — live. No surprise OpenAI bill at the end of the month.

  • 5
    Execution graph for full auditability

    Every tool call, every file write, every inter-agent message is logged in a visual node map. Debug complex multi-agent loops the way you debug a distributed system. Replay any branch. This is what makes Forge safe to run autonomously for 25 iterations while you sleep.

Forge vs The Competition

See how Forge stacks up against the leading AI coding tools. No other tool combines multi-agent collaboration, browser-native security, and zero-server architecture.

Feature Forge AI Claude Code Codex (OpenAI) OpenClaw Cursor NanoClaw
Multi-Agent Collaboration BEST
Browser-Based (No Install) CLI only CLI/IDE Desktop app
Sandboxed Plugin Execution UNIQUE
End-to-End Encrypted Sync AES-256
Extension Marketplace Limited
Local Models (Offline) Ollama + WebGPU
Self-Healing Linter UNIQUE
Shared Agent State (SQLite) UNIQUE
Execution Graph (Visual) UNIQUE
MCP Server Support Server + Client Client
Token Budget / Cost Tracking UNIQUE
Agent Evals Suite UNIQUE
Real-Time Multiplayer UNIQUE Limited
Autonomous Coding Mode
No-Code Builders UNIQUE
Zero Server / Zero Tracking UNIQUE Cloud Cloud Cloud
1-Click Deploy (Netlify/GitHub)
Price Free (BYO API keys) $20/mo+ $20/mo+ Free $20/mo+ Free
Open Source / Single-File YES

7 reasons Forge is the
best AI coding workspace

Other tools do one thing well. Forge does everything — securely, in your browser, with no servers.

01

True Multi-Agent, Not Just Chat

Claude Code and Cursor give you one agent. Forge lets you spawn multiple agents with different models — Claude for reasoning, GPT-5.5 for speed, DeepSeek for cost — all working on the same project simultaneously. They share state via SQLite, lock files to prevent conflicts, and you watch it all in the execution graph.

02

Zero Server, Zero Tracking

Cursor and Codex route your code through their servers. Forge runs 100% in your browser. Your API keys, your code, your conversations — they never touch a Forge server (because there isn't one). Cross-device sync is end-to-end encrypted with AES-256-GCM. Your keys never leave your device.

03

Sandboxed Plugins That Can't Hurt You

Every other tool either doesn't support plugins or runs them with full access to your system. Forge runs custom plugins in a sandboxed iframe — they can't read your API keys, localStorage, or cookies. Install marketplace extensions without fear. This is WebContainer-grade isolation, built in.

04

No-Code Bridge for Non-Devs

Claude Code requires a terminal. Cursor requires an IDE. Forge has visual form builders — non-devs create skills, plugins, and agents in plain English. "Make a plugin that fetches crypto prices" → working sandboxed plugin in seconds. Template mode lets non-devs build landing pages without writing code. Devs keep the raw code editor.

05

Cost Control Built In

Agent loops can burn through cash. Forge has a visual token budget — set a hard USD limit, and agents are force-stopped when they hit it. See cost per model, per agent, per request. Claude Code and Cursor give you no visibility into what you're spending. Forge shows you everything.

06

Offline-First, Not Cloud-Dependent

Claude Code, Cursor, and Codex stop working without internet. Forge supports local models via Ollama and WebGPU — run Llama 3, Mistral, or Phi on your machine. Auto-fallback when cloud APIs fail. Build on planes, trains, and in basements. Forge never stops working.

07

Deploy Without Leaving

Claude Code writes code — then you switch to a terminal to deploy. Cursor writes code — then you switch to Vercel. Forge has 1-click deploy to Netlify and GitHub built in. Self-healing checks run first. Push to GitHub with branch selection. Netlify goes live in seconds. Never leave the workspace.

Your code. Your keys.
Your device. Always.

Forge was built for environments where privacy is a hard requirement — defense, government, regulated enterprise, and any developer who refuses to ship their code to a third-party cloud. Because there are no Forge servers, your code and API keys literally cannot leave your browser.

  • API Keys Never Leave Your Device Stored in sessionStorage by default. Even with sync enabled, keys are NEVER included in the encrypted payload. Each device manages its own credentials.
  • End-to-End Encrypted Sync AES-256-GCM with PBKDF2 key derivation (100,000 iterations). Even if the sync backend is compromised, attackers cannot decrypt your data without your passphrase.
  • Sandboxed Plugin Execution Custom plugins run in isolated iframes with no same-origin access. They cannot read localStorage, sessionStorage, IndexedDB, cookies, or the parent DOM. Fetch is proxied with an allow-list.
  • XSS Protection All AI output is piped through DOMPurify before rendering. Live previews run in sandboxed iframes with noopener,noreferrer headers.
  • Zero Tracking, Zero Analytics No telemetry. No analytics. No accounts. Forge doesn't know who you are, what you build, or when you use it. Privacy by design.
Forge Security Model
// What stays on your device:
API keys → sessionStorage
Code files → localStorage
Agent state → IndexedDB
Conversations → localStorage
// What gets synced (encrypted):
Projects, agents, channels, plugins
// What NEVER gets synced:
❌ API keys (Anthropic, OpenAI, etc.)
❌ GitHub tokens
❌ Netlify tokens
Encryption: AES-256-GCM
Key derivation: PBKDF2 (100K iter)
Plugin isolation: sandbox iframe

An ecosystem, not a walled garden

Forge speaks Model Context Protocol on both sides — as a client that consumes external MCP servers, and as a server that exposes your workspace to the rest of your stack. This is what makes Forge a platform other tools plug into, not a tool you have to migrate away from.

Forge as MCP CLIENT

Connect your entire stack

Plug external MCP servers into Forge agents in one click. Your Architect agent can read a Linear ticket, fetch a Figma frame, query a Postgres replica, and post back to Slack — all from inside the same orchestrated loop. No glue code, no brittle integrations.

$ forge mcp add linear --scope architect
✓ connected · 42 tickets visible
Forge as MCP SERVER

Expose your workspace to anything

Forge itself runs as an MCP server. Your CI pipeline, your CLI scripts, your other AI agents — they can all call into Forge's project manager, file system, agent runtime, and deploy pipeline. Forge becomes a programmable engineering backend, not just an app you click on.

POST /mcp/tools/deploy_netlify
{ "branch": "main" } → ✅ live
GitHub
PRs · Issues
Slack
Notify · Approve
Notion
Docs · Spec
Drive
Files · Sheets
Figma
Frames · Tokens
Linear
Tickets · Cycles
Postgres
Query · Schema
Custom
Your MCP server

Procurement-ready.
Air-gap-capable. Fleet-managed.

Forge's zero-server architecture is not a limitation — it is the feature that lets it run inside defense networks, regulated enterprises, and government air-gaps where Cursor, Claude Code, and Codex cannot go.

Fully Air-Gappable

Self-host forge-bundled.html on an internal server with no outbound calls. Pair with local models via Ollama and Forge never needs to touch the public internet. Perfect for classified networks, SCIFs, and offline field deployments.

Single-Tenant Sync Backend

The cross-device sync backend can be deployed in your own VPC or on-prem. Your team's encrypted payloads never transit Forge infrastructure. Optional: fully disable sync and run each device as a standalone island with locally-managed keys.

Fleet License Management

Issue, rotate, and revoke license keys for entire teams from a single dashboard. Volume pricing for 50+ seats. Annual invoicing in USD or GHS with proper tax-compliant receipts — no per-seat credit-card friction for procurement teams.

Compliance-Ready Posture

Zero telemetry by design = nothing to disclose in a DPIA. No accounts = no PII to leak. AES-256-GCM + PBKDF2 satisfies encryption-at-rest requirements. DOMPurify + sandboxed iframes satisfy secure-execution controls. We provide a vendor questionnaire pack on request.

BYOK + Centralised Key Vault Option

Default: every developer brings their own provider keys, stored only on their device. Enterprise option: route all provider traffic through a central proxy with audit logging, rate limits, and cost allocation per project — without Forge ever storing the keys.

Procurement-Friendly Pricing

Pay in GHS, USD, or via mobile money. Annual invoicing, PO-based billing, and proper receipts for finance teams. No vendor lock-in: the open-source single-file build remains free forever, so a Forge deployment can never be held hostage by a billing dispute.

GDPR-friendly HIPAA-compatible SOC2-aligned Zero PII collected Air-gap deployable Single-tenant option Audit-loggable

Why the browser wins
over the desktop IDE

Desktop IDEs assume you own the machine, have admin rights, and don't mind a 600MB download. Forge assumes none of that. The browser is the new IDE — and it's the only place a truly secure, zero-install, BYOD-friendly AI platform can live.

Forge — Browser-Native

Runs anywhere a modern browser runs. That's the whole point.

  • Zero install, zero admin rightsOpen a URL and you're coding. IT never sees a software request.
  • BYOD & ChromeOS friendlyStudents, contractors, and locked-down enterprise laptops can all run Forge.
  • Instant updates, no restartsShip a fix at 9am, every user has it by 9:01. No updater binary, no version drift.
  • True sandbox by designThe browser's origin model + sandboxed iframes give defense-grade isolation for free.
  • Works on tablets & phonesReal cross-device, not "remote desktop to your real machine".
  • No background processesClose the tab = Forge is gone. No CPU-hogging indexers, no tray apps.

Desktop IDEs (Cursor, Codex, VS Code + extensions)

Powerful, but assume a world that fewer developers actually live in.

  • Installer + admin rights requiredBlocked on locked-down corporate, education, and government machines.
  • 500MB–2GB download per machinePainful on metered connections — common across Africa, Asia, and field deployments.
  • Plugin model = full filesystem accessA malicious extension can read your SSH keys. Desktop has no real origin isolation.
  • Version drift across teamEvery developer runs a slightly different build. Reproducibility suffers.
  • OS-specific bugsmacOS, Windows, Linux each ship their own quirks. Browser = one runtime.
  • Cloud-dependent for AIMost stop working offline. Forge falls back to Ollama + WebGPU.

Built in Accra.
Used everywhere.

Forge was built by George Lamptey (Gasby) in Accra, Ghana — a single-file, 20,000-line platform engineered to work on the kind of internet connection, hardware, and payment rails that most developer tools pretend don't exist.

That constraint became the product. Zero-server architecture means Forge runs fast on a hotspot. BYOK means a developer in Lagos pays the same token rate as one in London. GHS-native pricing means a student in Kumasi pays what they can actually afford — not a $20/month subscription designed for San Francisco salaries.

This is what African-built global developer tooling looks like: built for the realities of the majority world, useful enough that developers in Berlin, Bangalore, and Boston use it for the same reasons. Forge is proof that the next generation of foundational dev tools can come from anywhere — and increasingly, they will.

GHS 110
Local price / mo
~$7
USD equivalent
3
Mobile money rails
0
FX markup by Forge

Pay how you actually pay

Visa & Mastercard — every country
MTN Mobile Money (MoMo)
AirtelTigo Money
Vodafone Cash
Bank transfer (GHS)
Annual invoicing for teams

Powered by Paystack — African-built payment infrastructure. Your card or bank handles FX conversion at their standard rate. Forge never adds a markup.

"Forge isn't another AI code editor — it's an AI operating system for autonomous software engineering. Multi-agent orchestration, MCP-native extensibility, defense-grade privacy, and a browser-native runtime — built in Accra for the world."
George Lamptey (Gasby) · Creator of Forge AI · Accra, Ghana

One plan. Everything included.

No seat charges. No feature locks. All 36 modules, all providers — for the cost of a coffee a month.

Forge Access
$7/mo
GHS 110/mo · all cards accepted worldwide
F
All 36 modules — multi-agent, MCP, plugins, deploy, evals and more
7 AI providers — Claude, GPT, Gemini, DeepSeek, Groq, Ollama, WebGPU
Cross-device sync — AES-256-GCM encrypted, zero-knowledge
Priority updates — every new module ships to you first
100% BYOK — your AI keys, your costs, zero Forge markup
Email support — direct line to Gasby
Secured by Paystack  ·  Charged in GHS · all cards accepted  ·  No auto-renewal
Can I self-host Forge for free?

Yes — forge-bundled.html is free to self-host forever. The subscription covers the hosted workspace at theraventech.com/forgeai/, including backend sync infrastructure and all new modules the moment they ship.

What payment methods are accepted?

Paystack accepts Visa and Mastercard from any country worldwide, plus mobile money (MTN, Airtel, Vodafone) and bank transfers. Payments are processed in GHS — your card or bank handles the currency conversion automatically at their standard FX rate.

How does renewal work?

No auto-renewal. You pay for a period (monthly or annual), receive a license key by email, and renew when it expires. Forge notifies you 7 days before expiry with a renewal link.

Does the sub cover my AI API costs?

No — Forge is 100% BYOK. You pay Anthropic, OpenAI, etc. directly. The subscription is only for your access to the hosted Forge workspace. Forge never touches your tokens or adds markup.

Ready to build?
Start forging today.

Self-host for free forever, or subscribe for the hosted workspace with priority updates and backend sync from $7/month (GHS 110 · all cards accepted).

100% client-side  ·  BYO API keys  ·  Works offline