We can't find the internet
Attempting to reconnect
Something went wrong!
Attempting to reconnect
Forge runs your AI agents on self-healing infrastructure used by WhatsApp and Discord. When something breaks, it fixes itself in milliseconds — not minutes.
No credit card required. Cancel anytime.
Built on the same technology that powers
One crash kills everything. Like an ER with no walls — one patient codes, chaos spreads everywhere.
Shared process — failures cascade
Each agent is independent. Like a hospital with private rooms — one patient crashes, the others don't even notice.
Isolated processes — failures contained
For the technically curious
The BEAM is the Erlang virtual machine — originally built by Ericsson in 1986 for telephone switches that couldn't go down. It's the same runtime that powers WhatsApp (handling 100B+ messages/day), Discord (11M concurrent users), and Pinterest's notification system. Unlike Node.js or Python, the BEAM was designed from day one for fault tolerance, not just speed.
Every process in Forge runs under a supervisor — a parent process whose only job is to watch its children. If a child crashes, the supervisor restarts it automatically in microseconds. Supervisors can watch other supervisors, forming a tree. This means your agent can crash 1,000 times and each time it comes back instantly, with no human intervention. It's self-healing infrastructure, not just "we'll restart the container."
Docker containers restart in seconds to minutes. BEAM processes restart in microseconds. Docker gives you OS-level isolation; BEAM gives you process-level isolation within the same VM, so your agents share zero state. And unlike containers, BEAM processes are incredibly lightweight — you can run millions of them on a single machine. It's the difference between an apartment building (shared walls, shared plumbing) and true separate houses.
Sign up with your email. No credit card required.
Give it a name, a system prompt, and paste your LLM API key.
Your agent deploys at yourname.forge.anvilos.ai — ready to chat.
Powered by AnvilOS — the open-source AI agent operating system.
OTP supervisors restart crashed agents in microseconds with full context restored from the database. No pager alerts, no manual intervention.
Web chat, Telegram, Discord, and incoming webhooks. Extensible gateway architecture — add any platform with 5 callbacks.
Cron and interval scheduling built in. Your agents can check things, act on their own, and notify you — all defined in YAML.
Agents delegate, broadcast, and chain work to each other. Each runs in its own process with fault isolation — one crash never affects the team.
Conversations survive crashes and restarts. Each agent gets its own SQLite database on a dedicated machine — no shared state.
Use your Anthropic, OpenAI, or Google API key. Encrypted with AES-256-GCM. We never see your LLM traffic.
Not all AI hosting is created equal.
| Forge | Typical AI Host | |
|---|---|---|
| Runtime | BEAM VM (Erlang/Elixir) | Docker container |
| When an agent crashes | Supervisor restarts in <1ms | Manual restart or downtime |
| When one agent fails | Other agents unaffected | May cascade to others |
| Memory isolation | Per-process isolation | Shared OS memory |
| API key security | AES-256-GCM, per-machine DB | Shared database |
| Scheduled tasks | Built-in cron & intervals (YAML config) | External scheduler required |
| Agent-to-agent | Delegate, broadcast, handoff via OTP | HTTP calls between containers |
| Updates | Hot code upgrades, no downtime | Full redeploy required |
No hidden fees. No per-message charges. Cancel anytime. All prices in USD.
For solo builders getting started
For serious agent builders
For teams running production agents
Forge is managed hosting for AnvilOS — an open-source AI agent operating system built on Elixir/OTP. Think of it like WordPress.com for AI agents: you describe what your agent should do, and we handle the servers, deployment, uptime, and chat interfaces.
Most AI hosting runs on Docker containers with Node.js or Python. Forge runs on the BEAM — the same Erlang virtual machine that powers WhatsApp and Discord. This gives you automatic crash recovery in microseconds (not minutes), true process isolation between agents, and infrastructure that was designed 40 years ago for systems that can never go down.
Yes — Forge uses a Bring Your Own Key (BYOK) model. You provide your Anthropic, OpenAI, or Google API key. This means you control your LLM costs directly, and your key is encrypted with AES-256-GCM on your dedicated machine. Forge never proxies or sees your LLM traffic.
Each customer gets a dedicated machine with its own SQLite database on its own volume. There is no shared infrastructure between customers. Your conversations, agent configs, and API keys never touch another customer's machine. This is full physical isolation, not just multi-tenant database rows.
It restarts automatically. OTP supervisors monitor every process and restart crashed ones in microseconds — typically under 1 millisecond. Your agent's state is preserved in the database, so it picks up right where it left off. No alerts, no tickets, no waiting for someone to restart a container.
Yes. Agents can run on cron schedules or fixed intervals — checking things, acting on events, and notifying you through Telegram or Discord. They can also work as teams, delegating tasks to specialist agents and chaining results together. All configured in YAML, all running on OTP processes with full fault isolation.
Absolutely. AnvilOS is open source on GitHub. Forge is for people who want the convenience of managed hosting without dealing with servers, Docker, DNS, and SSL certificates. You can always export your agent config and run it yourself.
Your instance runs until the end of your billing period, then it's paused. Your data is retained for 30 days so you can reactivate anytime. After 30 days, data is permanently deleted. No lock-in, no exit fees.
Deploy your first agent in under 60 seconds. Self-healing infrastructure handles the rest.
No credit card required. Cancel anytime.