Foundation
A KNOWLEDGE_BASE.md every agent reads. Conventions, services, ADRs — one source of truth.
KNOWLEDGE_BASE.mdPowered by Anthropic's Claude — Requires API Key Authentication
Teams ship inconsistent code when AI runs without context. FORGE is a CLI that gives your Claude session what it's missing: a Knowledge Base, six specialist agents, and validation gates between phases. Run `forge plan`, `forge spec`, `forge implement`. Ship code your reviewers don't have to babysit.
Free Community tier · Requires an Anthropic API Key for tokensPrompts don't scale across a codebase. Agents don't either, on their own. What scales is structure — the team designs the information architecture, the AI executes within it.
Five principles that turn an AI into a teammate. Each pillar is a programmatic contract — not a prose checklist.
A KNOWLEDGE_BASE.md every agent reads. Conventions, services, ADRs — one source of truth.
KNOWLEDGE_BASE.mdThe Forge Lead decomposes features into tasks and coordinates six specialist subagents.
forge-lead → BA · SA · DBA · BE · FE · QAQuality gates with severity levels. TDD on every implementation. Code review before merge.
severity ∈ {blocker, major, minor}Reusable workflows that compose: forge plan → forge spec → forge implement.
$ forge {plan,spec,implement}Four validation levels — L1 syntax/compile, L2 unit tests, L3 integration, L4 full build — plus a confidence score from 1 to 10. Features don't merge below 7.
L1–L4 gates · exit 2 if score < 7The Forge Lead is present in all three phases — the connective tissue that holds the team together as agents come and go.
$ forge planUser stories, acceptance criteria, technical decisions, task list.
$ forge specComplete implementation blueprint: schema, contracts, test strategy.
$ forge implementProduction code with TDD. L1–L4 validation. Confidence score ≥ 7.
Real engineering roles with clear scope. The Forge Lead coordinates delivery; specialists execute their craft. Each command activates only the agents that phase needs.
An interactive cli renders agents, tool calls, and validation gates in real time. Three output modes: dashboard for humans, text for logs, JSON for piping.
The Knowledge Base is the single source of truth. Lean agents reference it instead of repeating it — context spent on signal, not noise.
# Each agent prompt You are a backend engineer. Conventions: - Use tsoa for routes - Use kysely for SQL - Use vitest for tests - …92 more lines of context Service map: - orders, billing, identity - …repeated for every agent Style guide: - …pasted into every prompt
# Backend Engineer role: implement server-side read: KNOWLEDGE_BASE.md read: PRP.md (for this task) output: - code with TDD - L1-L4 validation - confidence score ≥ 7 # That's the whole prompt.
FORGE agents are real subagents spawned via the Claude Agent SDK — not prompts pasted into a chat. The CLI authenticates against the Anthropic API using your own API Key.
# Install $ npm install -g @forge/cli # Authenticate with your Anthropic API Key $ forge login --api-key sk-ant-... ✓ api key validated · model: claude-opus-4-7 # Bootstrap your knowledge base $ forge init ✓ KNOWLEDGE_BASE.md created ✓ 7 agents registered
FORGE requires an Anthropic API Key to operate. This is the only supported authentication method.
Anthropic's terms of service require third-party tools to use API authentication, not consumer subscription credentials.
Pay-as-you-go pricing ensures transparent, predictable costs for your workflows.
API authentication is the officially supported method, guaranteed to work long-term.
# Install FORGE $ npm install -g @forge/cli # Login with your API Key $ forge login --api-key <your-api-key> # Or set as an environment variable $ export ANTHROPIC_API_KEY="sk-ant-..." $ forge init
Two clean line items in your bookkeeping: your FORGE plan + your Anthropic token spend.
See Anthropic's token pricing →You'll still need an Anthropic API Key to use FORGE. Subscription credentials cannot be routed through third-party tools — Anthropic enforces this server-side. Create an API Key, set up a payment method in the Console, and you're ready to go. This is the only compliant way to run FORGE, and we take Anthropic's terms seriously to protect both you and the integrity of this project.
“FORGE is built with compliance and longevity in mind. We use Anthropic's official API authentication because it's the right, sustainable way to build. This ensures FORGE works reliably for years to come — not just until Anthropic closes the next loophole.”
FORGE uses Anthropic's official API authentication methods exclusively. We do not support or encourage:
We believe sustainable AI tools are built on transparent, compliant practices. This benefits users, Anthropic, and the broader developer ecosystem.
Last updated: April 2026Two ways in: a free Community tier to test FORGE on real code, and a Pro subscription for serious projects. Anthropic token usage is billed separately, directly by Anthropic via your own API Key.
Perfect to test and learn FORGE on real code. Upgrade to Pro when you hit the limits.
No card required. You bring your own Anthropic API Key for token usage.
For teams and individuals shipping production code. Higher limits, multiple projects, longer retention.
Pricing announced before public launch. Early access subscribers get first look.
Both plans require an Anthropic API Key. You pay Anthropic directly for token consumption — FORGE never proxies your traffic.
Straight answers about authentication, costs, and how FORGE stays compliant.
We're onboarding alpha teams now. If you ship production code with Claude, get on the list.
Get notified when FORGE is open for early access. We'll send the install instructions and onboarding guide.
Free Community tier + your Anthropic API Key · Pro subscription pricing soon · v1 alpha