Get 10 hours back every week. AI agents that finish the work

AutoGPT is an AI agent platform: software for building, deploying, and running agents you can hand a goal to — they make a plan, then carry out every step using AI models and your everyday apps along the way. Unlike a chatbot that waits for your next message, an AutoGPT agent does the whole job itself: it can research a market, draft and send outreach, watch competitors, or keep a spreadsheet current, end to end.

You build agents whichever way suits you — describe the outcome in plain language and let AutoPilot assemble it, or wire the steps together in the visual no-code builder. Agents then run in the cloud on schedules and triggers, with every run visible in your dashboard. The platform is open source, with a hosted cloud for teams who would rather not run their own infrastructure — start from a marketplace agent or build your own from scratch.

You create an agent by describing the outcome you want in plain English, or by wiring up the steps in the visual builder — no code needed on either path. Under the hood an agent is a graph of blocks, where each block does one job: reading email, calling an AI model, scraping a page, updating a row in a sheet. AutoGPT connects those blocks to the models and apps the task needs and runs the work in the background.

Every run is recorded step by step, so you can open the dashboard and see exactly what an agent did and why. Agents run once on demand, on a schedule — say, every morning at eight — or on a trigger like a new email or form submission. Once it's set up, the agent keeps working without you watching, and you only step in when it needs a decision.

Anything that's repeatable work rather than a one-off — that's the sweet spot for no-code AI automation. Teams use AutoGPT agents to monitor the web and send an alert the moment something changes, draft and send outreach, research prospects before a sales call, turn long documents into clean summaries, generate content briefs, triage a support queue, and keep spreadsheets and CRMs up to date without anyone touching them.

You don't have to start from a blank canvas either: the marketplace is full of ready-made agents you can deploy in a click and then tweak to fit how you work. A good rule of thumb — if you can describe the job clearly enough to hand it to a capable assistant, you can build an AI agent for it, and the agent will keep doing it on whatever schedule or trigger you choose.

A contractor is the right call for judgement-heavy, one-off work — a brand redesign, a tricky migration, advice you'll act on once. AutoGPT is built for the repeatable kind of work that comes back every day or every week. An agent runs around the clock, handles many tasks in parallel, responds in seconds rather than days, and costs a fraction of an hourly rate, because you only pay for the compute and model usage each run actually consumes.

Agents also don't context-switch or take holidays: the daily briefing lands every morning and the competitor watch never blinks. They're not a replacement for human judgement on novel, high-stakes decisions — they're leverage for everything that happens again and again. Most people end up using both: a person for the bespoke work, and a handful of agents for the recurring tasks that used to eat that person's week.

They solve different problems. Claude Code and Codex help you write code, and Perplexity answers questions with sources — they assist you, in the moment, while you're actively prompting them. AutoGPT is an AI agent platform: you build agents that run on their own, across your apps, on a schedule or a trigger, doing the whole job end to end rather than the part you're working on right now.

Put differently, an assistant helps you draft one reply; an agent clears the queue while you're in a meeting. The two are complementary — plenty of AutoGPT users still reach for a coding assistant or an answer engine during the day, then let agents own the recurring work. For a side-by-side look at how AutoGPT stacks up against automation tools and agent frameworks, see our comparison pages covering Zapier, n8n, LangChain, and more.

Yes, if you're technical: AutoGPT is open source, so you can self-host the whole platform for free on your own infrastructure — you bring your own model API keys and hardware, and there's no licence fee or seat limit. Everything you build there maps over to the cloud later if you switch.

The hosted cloud doesn't have a free tier, and that's deliberate: every agent run uses real compute and paid AI-model usage that we cover on your behalf, so a free plan would mean throttling agents into uselessness. Instead, each plan includes enough usage to put agents to work from day one — AutoPilot Chat, the visual builder, the marketplace, and a credit wallet for your agents' runs. If you want to kick the tyres before paying, self-hosting is the genuinely free path — and the open-source repo is the same codebase the cloud runs.

Running agents is pay-as-you-go: each run draws from a pre-paid credit wallet you can top up any time, and the rate is the same on every plan — you never pay a premium per task for being on a smaller tier. Usage is what your agents actually consume: the AI-model calls they make, the compute their blocks use, and any premium integrations they touch along the way.

Light agents cost very little per run; heavy research agents cost more, and the dashboard shows exactly what every run consumed, so there are no surprises at the end of the month. Your subscription (Pro or Max) sets how much AutoPilot Chat usage is included — that's the main difference between the plans. In short: the subscription covers chat, the credit wallet covers the agents you run, and the pricing page has the full breakdown.

Yes. AutoGPT is open source, and you can run the entire platform yourself with a single Docker Compose command — on Linux, WSL, Podman, even a Raspberry Pi. Self-hosting is free: you bring your own model API keys and infrastructure, and you get the same visual builder, block library, and agent runtime the cloud runs.

It's the right choice when data can't leave your network, when you want full control over models and costs, or when you simply like running your own stack. The hosted cloud is the managed, zero-setup alternative — we run the servers, cover model usage through your plan, and keep everything up to date. Plenty of teams prototype in the cloud and move sensitive workloads on-prem later, or start self-hosted and graduate to the cloud as they scale. Start from the GitHub repo, or follow the self-hosting guide in the docs.