The easiest, most extensible way to build an agent you'll actually rely on — instructions and behaviors independent of any single model, on infrastructure built to be trusted.
14-day money-back guarantee. If it's not for you, full refund — and we hold your export for 7 days.
Your agent runs on its own managed instance — not shared infrastructure.
Automatic daily backups with point-in-time restore. Never lose work.
Every session, every tool call, every token — tracked and searchable.
Set up out of the box — message your agent the moment it's live. Add WhatsApp, Teams, or Google Chat in a few clicks. No command line.
Pre-built configurations for common roles. Customize to fit your workflow.
Download your agent's full workspace and run it on your own infrastructure.
Built on our open-source agent framework. Skills, memory, and tool use out of the box.
Private forum for sharing configurations, tips, and use cases with other operators.
Pick a name for your agent and select a plan. Choose a pre-built template or start from scratch.
Secure checkout via Stripe. Your dedicated server is automatically provisioned in the background.
Open your dashboard, connect your tools, and start messaging your agent immediately.
The cheapest way in — and no way to run a system you rely on. An unattended agent holding credentials to your accounts, on your home network, with no monitored security, no isolation, backups if you remember, and consumer-grade uptime. A fine weekend project. Not an employee.
Easy to start — and everything you build is stuck there. Your instructions, behaviors, and workflows are written against one company's models, one company's pricing, and one company's features. When they change, you go along.
Build instructions and behaviors independently of any model. Run them on a dedicated, managed server with monitored security, backups, and a full audit trail. Choose from vetted models — swap anytime, pin anytime, export anytime. This is the extensible, future-proof way to run an agent your business depends on. The pricing reflects a system built to be relied on.
A steady part-time assistant. ≈55–70 routine actions a day — email triage, scheduling, research summaries — with deep-reasoning tasks when you need them.(interim estimate)
A full-time workhorse. ≈150–250 routine actions a day plus daily deep work — connected to your real business systems.(interim estimate)
Maximum sustained throughput. ≈400+ routine actions a day, heavy deep-reasoning workloads, multiple concurrent workflows.(interim estimate)
Best value for raw horsepower. Your agent runs on the ChatGPT subscription you already pay for; we provide the managed platform around it.(interim estimate)
Deep-reasoning models burn credits faster than routine models — one currency, your choice of where to spend it. See what's under the hood ↓
Quick, high-volume actions — triage, lookups, short replies — at a quarter of a credit's burn.
The everyday workhorse class for most agent tasks — the baseline a credit is measured against.
Heavy multi-step reasoning and planning — the most capable models, for when the task is worth it.
Every model here has passed our tool-calling reliability evals on Hermes. If it's not on this list, it's not because we're hiding it — it's because it failed.
Upgrade policy. Auto-upgrade within class is the default — when a better model clears our evals, your agent gets it. Prefer stability? Pin your model. We'll never switch what's running your agent without telling you, and pinned models get 60–90 days' notice before retirement.
A 30-minute working session, booked at least 2 business days out. For planned work — new integrations, workflow changes, a second pair of hands.
A 30-minute session with a response within 1 business day. For when something needs attention now.
2 sessions monthly, roll over one month, cancel anytime. For builders who want a standing pair of hands.
Roughly one routine agent action. Deep-reasoning work burns faster. Your dashboard shows exactly where they go.
Your agent never stops mid-task. It shifts to fast, efficient models and you can top up in one click — or wait for your monthly reset.
A burst ceiling of ~3× your daily average protects against runaway loops. Normal use never touches it.
Monthly credits don't; top-ups last 90 days.
Connect via OAuth in your dashboard; your agent's inference runs on your subscription; FxM provides the server, security, backups, audit trail, templates, and community.
An active ChatGPT Pro $100 (5x) or $200 (20x) plan. Your provider's rate limits govern throughput — the $200 plan is the best token value on this page, and what we'd recommend for a full-time agent.
If your subscription hits its limits mid-task, your agent can fall back to FxM platform credits so work never stalls. You control this in settings.
If you already pay for ChatGPT, this is the best value here. We'd rather tell you that than hide it.
Try it for two weeks. If it's not for you, request a full refund of your first payment — self-serve, no questions. We pause your agent and hold your export for 7 days so you keep everything you built. Read the refund policy →
Name it, pick a role, and you'll be up and running in minutes.