Help & Support

Connect your tools, describe what you want in plain English, and the agent is ready to run. No coding required.

On Pro and up you choose, per agent, from the models we run. Free agents run on Auto, a cost-effective model we choose. It's a dropdown, not a migration: the same agent, tools, knowledge, and autonomy dial work whichever model you pick, so you're never locked to one vendor's roadmap.

Auteryn is where you configure agents that do the work: each gets its own persistent machine — terminal, browser, files — and runs in loops until a separate check agrees the goal is met, or your budget says stop. You watch every pass live and can take over its browser mid-run. Agents connect to your real tools, and answer your customers on their own channels. Our ChatGPT comparison (auteryn.ai/compare/chatgpt) sets the pricing side by side.

Yes — continuing work is built in. Agents run on schedules and event triggers (a webhook, a customer message, a build finishing), book their own follow-ups (serial chains, three pending at most), and work toward goals where a separate check — not the agent grading itself — decides when it's done. Background jobs wake the agent when they complete, and interrupted runs resume with plan, notes, and tool history intact. Every pass is a normal run you can watch live and step into.

Loop engineering is the industry's June-2026 name for designing the system around an agent — triggers, verification, budgets, and stop rules — instead of prompting it by hand. Our take: loops are only trustworthy when they're bounded and visible. Auteryn loops carry run budgets, checkpoint near the limit, stream every step to a live timeline, and keep a human seat — you can step into any pass.

Yes — structurally. After qualifying runs, a reflection pass stores what worked and what didn't, and the next similar task starts with those lessons. Agents also keep self-evolving long-term memory: they write and consolidate their own notes, searchable in natural language — and you can read and edit every entry. When an agent notices a workflow it keeps repeating, it can draft that workflow as a new Skill; every self-drafted Skill waits in a proposed rail for your approval before it can run anywhere.

Yes, on paid plans. An agent can hand parts of a wide job to helpers (up to 4 per fan-out on Pro, 8 on Business and 12 on Enterprise, two of them working at a time). They share the agent's computer, you watch each one work in the chat, and the agent writes one answer from what they return.

Yes. Describe what you want and Studio researches your own catalog and drafts the agent or Skill, streaming its progress as it works. You can also hand it a chat thread where the workflow already happened, a screen recording of the task, or a YouTube link. Drafts always land in the editor for your review — never straight to production.

Budgets are structural, not advisory. Every run carries hard time and model-call limits; near the budget the run checkpoints its state cleanly instead of overrunning. Credits are transparent — 1 credit = $0.001, forever — every action is metered and visible, and free-plan runs are additionally capped at 150 credits per run.

Agents only access what you explicitly connect. Tokens are short-lived and scoped, fetched just-in-time for each call, and revoking a connection stops access immediately. Credentials are encrypted at rest with AES-256-GCM, and traffic runs over TLS.

You're always in control. An agent's autonomy level sets when it stops to ask you: Co-pilot asks before every change, and Autopilot asks only before payments and credentials. Every tool call is in the run timeline, so you can see exactly what happened.

No. Your data is never used to train AI models. All interactions are private to your organization.

Each agent works on its own computer. When a job is wide, it can hand parts of it to helpers that share its computer, and it writes one answer from what they return (helpers are on paid plans). Within a run it can also call focused sub-agents for research, review and computer use, and when you set a goal, a separate check reviews each pass.

Computer Use is our vision-based UI automation capability. Agents can see, click, type, and navigate any screen just like a human — enabling automation of any web application without APIs.

For days or weeks. Work happens in runs that are checkpointed after every step, and the task carries on from one run to the next because the loop continues — schedules, events, follow-ups the agent books, and goal passes — with sessions, files, and context persisting between runs. Interrupted runs resume with full state.

It pauses, and the next time the agent needs it, it resumes where it left off: its open browser pages and running commands are still there, along with its files. Connections to other servers, such as an SSH session, close during the pause and need to reconnect. If a computer can't resume this way, it starts fresh with its files put back. This applies to computers in our cloud, not to your own machine.

Auteryn uses a unified credit system — one credit balance covers AI conversations, agent actions, media generation and voice. 1,000 credits = $1, always. Start free with 1,000 credits/month. Paid tiers add more of everything — Pro $29 with 15,000 credits and 25 computer hours, Business $99 with 60,000 pooled credits and 100 computer hours — plus top-up credit packs that never expire. Computer time is a monthly allowance of hours, not a credit charge: when the hours run out, new computers wait until next month or an upgrade.

The Free tier includes 1,000 credits, 1 computer hour, and 100 MB of storage per month — enough to build and test your first agents. Upgrade any time when you need more capacity.

Credits cover AI and media work: conversations (about 5–40 for a quick chat, roughly 50–500 for a deep multi-step run), images (50–95 each, 340 at 4K), video (200–550 per second, Pro and up), voice conversations (30/min in the browser, 54/min by phone), speech (50–60 per 1,000 characters), and transcription (10/min). Knowledge queries and Loop triggers are free. Computer time is a monthly allowance, not a credit charge. Once the hours are used, new computers wait until next month or an upgrade. Your monthly allocation covers most use cases. Pro and Business can add top-up packs, which never expire.

Yes — two ways. On any paid plan you can bind an agent to a Linux machine you own: one command installs a signed runner, the agent's shell and files run on your hardware, in the folder you choose. Separately, an Enterprise contract can add a dedicated AWS account with customer audit role access, VPC peering, and BYOK (bring your own keys). Contact us for custom infrastructure requirements.