Skip to main content
Coming Soon

From prompt to
pull request.

Queue tasks, assign AI agents, watch live terminal output, and review auto-created PRs. Forge turns Claude Code into a managed fleet of specialized developers.

Quality Gates Git-Native Real-Time Streaming
Get Early Access on WhatsApp See How It Works

How It Works

Four steps to shipped code

1

Create Task

Describe the work in plain English. Set priority and target module. That's it.

2

Assign Agent

Pick a specialized agent — Coder, PM, DocWriter, or Marketer. Each has its own persona and model.

3

Watch Live

Real-time terminal streaming via WebSocket. See every file read, edit, and command as it happens.

4

Review PR

Forge auto-creates a branch, commits changes, pushes, and opens a GitHub PR — ready for your review.

forge — OPHQ Coder — FG-2026-0042

$ forge run --task FG-2026-0042 --agent "OPHQ Coder"

▸ Reading api/internal/seva/store.go...

I'll add pagination to the ListPatrons query using

LIMIT/OFFSET with a cursor-based fallback.

▸ Editing api/internal/seva/store.go

▸ Editing api/internal/seva/handler.go

▸ Running go build ./...

✓ Build passed (2.1s)

▸ Running go vet ./...

✓ Vet passed (0.8s)

▸ Running go test ./...

✓ Tests passed (12.4s)

▸ git push origin forge/FG-2026-0042-add-pagination

✓ PR created: github.com/annailabs/ophq/pull/142

Completed in 45.2s · 1,234 input tokens · 567 output tokens

Features

Everything you need to manage AI agents

Forge isn't another chatbot wrapper. It's infrastructure for running a fleet of autonomous coding agents.

Agent Fleet Management

Create specialized agents with distinct personas, models, and tool permissions. Track per-agent success rates, token usage, and average run times.

Priority Task Queue

Organize work by priority — critical, high, medium, low. FIFO within each tier. Agents pick up the most important work first, automatically.

Quality Gates

Automated build, vet, and test checks after every run. Add custom gates for linting or security scanning. Nothing ships without passing.

Git Integration

Auto-create branches, commit changes, push to remote, and open GitHub PRs. Every task gets its own branch — no conflicts, clean history.

Real-Time Terminal

WebSocket streaming of every agent action — file reads, edits, bash commands, and reasoning. Tokyo Night themed terminal UI in your browser.

Autonomous Mode

Enable agents to auto-pickup pending tasks from the queue. They poll, execute, gate-check, and ship — 24/7, no human in the loop.

Agent Roster

Specialized agents, purpose-built roles

Forge ships with 4 default agents. Create custom agents for any workflow.

Agent Model Persona Use Cases
Coder
claude-sonnet-4-6 Full-stack developer. Follows project patterns, writes clean code. Features, bug fixes, refactoring, API endpoints
PM
claude-sonnet-4-6 Product manager. Writes clear specs and user stories. PRDs, specs, user stories, acceptance criteria
DocWriter
claude-haiku-4-5 Technical writer. Keeps docs in sync with code. API docs, READMEs, changelogs, feature guides
Marketer
claude-sonnet-4-6 Growth marketer. Creates landing pages and SEO content. Landing pages, blog posts, SEO copy, email campaigns

+ Create unlimited custom agents with any role, model, and system prompt

Deep Dive

How AI agent orchestration works

AI coding assistants are powerful one-at-a-time. But real development teams work in parallel, with specialists handling different concerns. Forge brings that model to AI.

Task Lifecycle

Pending Running Quality Gate Review Done

Every task moves through a defined lifecycle. Failed quality gates automatically retry before escalating.

Quality Gate Pipeline

go build ./...

Does the code compile? Catches syntax errors, missing imports, type mismatches.

go vet ./...

Static analysis. Finds suspicious constructs, unreachable code, incorrect printf args.

go test ./...

Run the full test suite. Ensures existing tests still pass after the agent's changes.

Forge vs. alternatives

Capability Forge CI/CD Pipelines Chatbot Wrappers
Writes code from natural language Yes No Yes
Automated quality gates Yes Yes No
Git integration (branch, commit, PR) Yes Partial No
Specialized agent personas Yes No No
Task queue with priorities Yes Partial No
Real-time output streaming Yes Partial Yes
Autonomous 24/7 operation Yes No No

FAQ

Common questions

What AI models does Forge use?
Forge uses Anthropic's Claude models through the Claude Code CLI. You can assign different models per agent — Claude Opus for complex architecture work, Claude Sonnet for everyday coding, and Claude Haiku for documentation and quick tasks. Each agent is configured independently.
How does the quality gate system work?
After every run completes, Forge automatically runs a pipeline of quality checks: go build (compilation), go vet (static analysis), and go test (unit tests). Each gate must pass before the run is marked complete. You can also add custom gates like linting or security scanning. If any gate fails, the run is marked failed with detailed output.
Is my code secure with Forge?
Yes. Forge uses multi-tenant isolation with PostgreSQL row-level security. Each business's data is completely separated. API keys are stored encrypted and never exposed in responses. Agent subprocesses run in isolated OS processes with configurable tool restrictions, timeouts, and resource limits.
What's the pricing model?
Forge uses a BYO (Bring Your Own) API key model. You provide your Anthropic API key and pay Anthropic directly for Claude usage. There is no platform fee during early access. This keeps costs transparent — you only pay for the AI compute you actually use.
Can I customize agents?
Absolutely. Every agent is fully customizable. You can set the role, Claude model, system prompt, maximum turns, allowed tools, and concurrency limits. Create agents for any workflow — a security auditor, a test writer, a documentation specialist, or any custom role you need.
Does Forge work with any programming language?
Since Forge runs Claude Code under the hood, it supports any language that Claude Code supports — including Go, Python, JavaScript, TypeScript, Rust, Java, and more. The built-in quality gates are configured for Go projects by default, but you can add custom gates for any language's toolchain.
What happens when a run fails?
When a run fails (either from a subprocess error or a failed quality gate), the full error output is captured and stored. You can review the terminal output, see exactly which gate failed and why, then retry the task. Forge supports auto-retry with configurable max retry counts.
Can agents work autonomously without human input?
Yes. Agents with autonomous mode enabled will automatically pick up pending tasks from the priority queue and execute them. They poll for new work on a configurable interval (default: 30 seconds). Combined with quality gates, this means agents can continuously ship tested code 24/7.

Ready to automate your
development workflow?

Forge is in active development. Sign up for early access and be the first to deploy a fleet of AI coding agents on your codebase.

Chat With Us on WhatsApp