ForgeOS
The governance OS for agentic software teams
Overview
ForgeOS is the governance OS for teams running agentic software pipelines. It is not an AI — it is the persistent enforcement, memory, and trust layer that sits between your AI agents and your infrastructure.
Every commit, every deploy, every inter-agent delegation passes through a configurable gate. Gates require artifacts — architecture approvals, QA sign-offs, security reviews — before the next phase unlocks. No artifact, no execution.
The audit ledger is Ed25519-signed and hash-chained. Every action is attributed, ordered, and tamper-evident. When something goes wrong, you trace it in seconds, not hours.
SharedMind is the institutional memory layer — patterns, anti-patterns, and lessons persist across sessions and are surfaced automatically to any agent that connects. Your AI agents get smarter every cycle, without manual curation.
Ships as an MCP server (21 tools), a CLI (forge — 18 command groups, 49 subcommands), and a REST API (239 endpoints). Works with any MCP-compatible agent: Claude, GPT, Gemini, or local models. Connect via MCP with your ForgeOS API key — your AI model keys stay with you. Raw source code stays local — ForgeOS analyzes project metadata (file structure, languages, dependencies) but never uploads file contents.
WHY FORGEOS
Enforce, don't observe
Gates fire before actions execute. Policy is constitutional - defined once, enforced everywhere. No configuration can disable self-approval prevention.
Immutable audit trail
Ed25519-signed, hash-chained ledger. Every agent action, every gate decision, every delegation event. Retroactive alteration breaks the chain - detectable in milliseconds. Auditor-ready. Designed for EU AI Act readiness.
Agent-native + CLI
21 governance tools over MCP (HTTP/SSE). Any MCP-compatible agent calls ForgeOS directly -
gate-check, delegation-validate, audit-query. Plus the forge CLI for terminal-first workflows. Framework-agnostic. No SDK required. Works with Claude, GPT, Gemini, or local models - your AI keys stay with you.
Features
Enforcement gates
Every commit, deploy, and delegation passes through configurable governance gates before execution.
Ed25519-signed audit ledger
Hash-chained, cryptographically signed event trail. Every action is attributable and tamper-evident.
MCP server (21 tools)
Drop ForgeOS into Claude Desktop, Cursor, or any MCP-compatible agent. Works with Claude, GPT, Gemini, and local models.
forge CLI
18 command groups, 49 subcommands. Run gate checks, query SharedMind, manage projects, and inspect the trust ledger directly from your terminal.
SharedMind
Institutional memory that persists across sessions. Patterns, anti-patterns, and lessons are stored and surfaced automatically — your AI agents learn from past work.
Federation
Protocol-based interfaces let multiple teams and products share governance infrastructure. Improvements propagate across all layers.
Stripe billing built in
Subscription billing with AI spend caps, enforced at the API layer. No payment, no access.
TRUST SIGNALS
Governing 13 AI departments in production - including the team that built it.
Dogfooded in production
ForgeOS governs the development of ForgeOS. Thirteen specialized AI departments - engineering, security, quality, operations, and nine others - operate under ForgeOS governance every day. Real agents. Real violations blocked.
The 41% problem
41% of official MCP servers currently lack authentication, based on our analysis of public MCP server listings (SyncTek, 2026). ForgeOS is an MCP-listed governance tool with authentication enforcement by default - not as a configuration, as a requirement.
EU AI Act - enforcement begins Aug 2, 2026
ForgeOS maps directly to Article 12 (record-keeping) and Article 17 (quality management system). Enterprise procurement cycles run 90+ days. Evaluation must begin by May 2026.
EU AI Act readiness: what ForgeOS covers
| Article | Requirement | ForgeOS coverage | Coverage level |
|---|---|---|---|
| Art. 12 | Record-keeping - immutable logs of all automated decisions | Hash-chained, Ed25519-signed ledger | Direct coverage |
| Art. 9 | Risk management system - documented procedures | Gate enforcement, circuit breakers, violation scanner | Direct coverage |
| Art. 14 | Human oversight - ability to halt and override | Human-in-the-loop gates, kill switch, circuit breaker | Direct coverage |
| Art. 17 | Quality management system | Initiative + gate lifecycle | Direct coverage |
| Art. 61 | Post-market monitoring | Continuous automated scanning, daily self-audit | Partial coverage |
ForgeOS provides technical infrastructure that supports EU AI Act compliance. Consult legal counsel for your specific conformity assessment requirements.
CATEGORY DEFINITION
What is agentic workflow governance?
Agentic workflow governance is the discipline of defining, enforcing, and auditing what AI agents are permitted to do in a multi-agent system. It is distinct from model governance (which governs AI model behavior) and agent observability (which records what agents did after the fact). Governance enforces rules before actions are taken - through gates that require human approval, delegation rules that define agent authority, circuit breakers that halt execution on failure, and self-approval prevention that blocks agents from authorizing their own outputs. ForgeOS is the first purpose-built governance OS for MCP agents.
Where ForgeOS fits in the stack
| Layer | Tools | ForgeOS relationship |
|---|---|---|
| Code generation | Cursor, Windsurf, Devin | They produce code. ForgeOS governs what happens to it after. |
| Agent orchestration | CrewAI, LangGraph, AutoGen | They route tasks. ForgeOS enforces gates, delegation rules, and audit trails. |
| Observability | LangSmith, W&B Weave | They record what happened. ForgeOS prevents what should not happen. |
| ForgeOS | ForgeOS | The enforcement layer. Between orchestration and human approval. |
MintMCP secures the connection. LangSmith records the trace. CrewAI routes the task. ForgeOS enforces the rules - before the action is taken, with a cryptographic receipt that proves it.
AGENT DISCOVERY
Your agents can find us.
ForgeOS is designed to be discovered and called natively by AI agents - no SDK, no custom integration, no human intermediary required.
Model Context Protocol
ForgeOS exposes 21 governance tools over MCP. Any MCP-compatible agent - Claude, GPT-4, Gemini, local models - can call gate-check, delegation-validate, and audit-query without modification. Works with any MCP-compatible AI - Claude, GPT, Gemini, or local models. Your AI keys stay with you.
Agent-to-Agent discovery
ForgeOS issues A2A Agent Cards - machine-readable capability manifests. Any A2A-compatible agent can discover, evaluate, and interoperate with ForgeOS autonomously. No human required to wire the integration.
LLM-readable discovery
ForgeOS publishes an llms.txt
manifest for LLM-native documentation discovery. Agents browsing for governance tooling find ForgeOS without a human intermediary.
Pricing
Free
$0
1,000 API requests/month, 20 req/min — get started with ForgeOS governance
Pro
$49/mo
50,000 API requests/month, 100 req/min, 10 AI reviews/min — full enforcement suite
Team
$29/seat/mo
3-seat minimum. 200,000 API requests/month. Shared projects, role-based access.
Enterprise
Custom
Unlimited requests, dedicated tenant, SLA, DPA, security questionnaire
Honest tradeoffs
- Commercial hosted service — the CLI is MIT-licensed, but the ForgeOS API requires a paid subscription.
- Requires integrating gate checks into your agent workflow.
- MCP server requires a running ForgeOS instance (cloud-hosted).
- Full enforcement mode blocks non-compliant actions — intentional, but plan for it.
- Cloud-connected — governance operations require network access to the ForgeOS API. No offline mode.
- Rate limits apply per plan tier — higher tiers unlock more API throughput for larger teams.
Documentation
Support
Related posts
We Ran 25 Packages Through 3 Scoring Tools. Here's What We Found.
Cross-tool comparison of OpenSSF Scorecard, Registry Quality signals, and ForgeOS Trust Index across 25 popular open-source packages. Raw data included.
The 3-Week Wall: Why AI Agent Workflows Degrade Without Governance
Almost every team using AI coding agents hits the same wall around week three. The pattern is predictable -- and so is the fix.
Why AI Agent Workflows Need Governance -- And What That Looks Like
AI coding agents are fast. But without gates, artifacts, and audit trails, speed becomes chaos. Here's what governance means for agentic workflows.