What is AutoGPT?
AutoGPT began in 2023 as a single Python agent that looped on its own goals. The repository now centres on the AutoGPT Platform: a web app where a coordinator called Otto hands work to role-based agents (an SEO writer, a sales researcher, an executive assistant), backed by a visual workflow builder, a marketplace of shared workflows, scheduled routines, an API and an MCP server. The original agent, the Forge toolkit and the agbenchmark suite live on under classic/.
The licence is split. Everything inside autogpt_platform/ is under the PolyForm Shield 1.0.0 licence, which allows personal and internal business use but forbids offering it as a competing product; everything outside that folder, including classic/, is MIT. Significant Gravitas sells the hosted Platform on Pro and Max plans; self-hosting runs from one Docker image with your own model keys.
Is AutoGPT worth it?
The name still carries more weight than the software. With about 188,000 GitHub stars it remains one of the most-starred AI repositories, but the README now reads as a landing page for the hosted product, with sign-up links throughout and a cast of named agents. Releases are frequent: platform beta v0.11.0 shipped on 4 November 2026 and v0.11.2 on 11 November, and commits land daily.
The licence decides whether it fits. The platform code is PolyForm Shield, not an OSI-approved open-source licence: a company can run it internally, but cannot offer it to customers as a service that competes with AutoGPT's own. Only the classic agent and the surrounding tooling are MIT. Teams comparing it with n8n or Dify should know that both of those also carry restrictions, but that Langflow and CrewAI do not.
Self-hosting has become simpler and still needs care. The single-container image binds to 127.0.0.1:3000 by default, and the self-hosting guide reports about 5 to 6 GiB of memory in testing. New-account creation is open by default, so anyone who can reach the port can register until you set an allowlist or switch sign-ups off; email verification is not supported. Chat runs through OpenRouter by default and memory needs an OpenAI key for embeddings as well; spend caps are disabled unless you set them, and backups are cold only, written as plaintext archives that contain secrets.
The security record is the main concern. The repository has published 45 advisories in 2026, 10 rated critical and 17 high, among them a hard-coded default encryption key that exposed stored OAuth credentials, an SSRF filter that the address :: slipped past, and a Sentry tunnel that forwarded session cookies. Fixing them is to the team's credit; the count says the platform holds credentials for many connected apps and has a large attack surface.
Our take: for a team that wants a hosted agent to draft content and research prospects, the $50 Pro plan is a fair trial, but credits for scheduled runs are billed on top and the included allowance is not stated. For developers building their own agents, a framework with a plain MIT or Apache licence is the safer foundation.
Where it earns its place
A three-person marketing agency
Turn a target keyword into a brief and a draft article every week
Brief the content Expert on the hosted Platform and set it as a routine. Pro is $50 a month after the 7-day trial; scheduled runs draw on the separate credit wallet, so set a per-Expert spending cap before leaving it running.
An IT lead at a 200-person firm with a no-SaaS rule for client data
Run internal agents on company servers with company model keys
Run the single Docker image behind a TLS reverse proxy, set AUTH_SIGNUP_ALLOWLIST to the company domain and turn new accounts off after the admin is promoted. Budget at least 6 GiB of memory and encrypt the cold backups, which hold secrets in plaintext.
A developer wiring agents into an existing product
Trigger saved workflows from their own code or a coding agent
Call workflows through the Platform API or its MCP server. Internal use is allowed under PolyForm Shield, but reselling the platform as a service is not, which matters before it goes into a customer-facing product.
Before you commit
- Setup
- `docker run` with the significantgravitas/autogpt image, wait a few minutes for first boot, open localhost:3000, register and promote yourself with `autogpt-admin promote`; an installer script wraps the same steps.
- Integrations
- OpenRouter, OpenAI, Anthropic, Ollama and other OpenAI-compatible servers; connected apps for each agent; a REST API and an MCP server; a marketplace of shared workflows.
- Data and privacy
- Self-hosted data stays in the /data volume, with backups as plaintext archives that include secrets. The hosted pricing page lists SSO, ClamAV scanning of uploads and audited admin impersonation, but says nothing about retention or training use.
- Cost at scale
- Hosted Pro is $50/month and Max $320/month with 8.5 times Pro's usage; workflow runs are pay-as-you-go from a credit wallet, and yearly billing saves 15%. Team pricing is not published.
- Limits
- About 317 open issues. Self-hosting has no hot backups, no in-place secret rotation and no CLI password reset; spend caps are off by default; uploads top out at 256 MB on the hosted service.
What’s good
- Visual workflow builder, marketplace and scheduled routines in one app
- One-command Docker self-hosting, bound to localhost by default
- Works with OpenRouter, Anthropic or local models via Ollama
- Very active: platform releases several times a month
What’s not
- Platform code is PolyForm Shield, not open source
- 45 security advisories in 2026, 10 critical
- Self-hosted sign-up open by default; no email verification
- Hosted credit allowance for workflow runs not published
Best for
Small business teams that want a hosted agent for recurring content, sales and admin work and accept a source-available licence.
AutoGPT alternatives
AutoGPT: common questions
- Is AutoGPT worth it?
- AutoGPT is now a commercial agent platform with a famous repository attached, and its licence, its open sign-up default and a heavy 2026 run of security advisories all argue for reading the small print first. We rate it 3.5 out of 5.
- How much does AutoGPT cost?
- Self-hosting free (platform under PolyForm Shield, the rest MIT); you pay for infrastructure and model usage. Hosted Pro $50/month with a 7-day trial, Max $320/month; workflow runs billed from a separate credit wallet. Team plan not yet priced.
- What are the downsides of AutoGPT?
- Platform code is PolyForm Shield, not open source; 45 security advisories in 2026, 10 critical; self-hosted sign-up open by default; no email verification; hosted credit allowance for workflow runs not published.
- Who is AutoGPT best for?
- Small business teams that want a hosted agent for recurring content, sales and admin work and accept a source-available licence.
- How much does AutoGPT cost for a team or in production?
- Hosted Pro is $50/month and Max $320/month with 8.5 times Pro's usage; workflow runs are pay-as-you-go from a credit wallet, and yearly billing saves 15%. Team pricing is not published.
- Is AutoGPT safe for company data?
- Self-hosted data stays in the /data volume, with backups as plaintext archives that include secrets. The hosted pricing page lists SSO, ClamAV scanning of uploads and audited admin impersonation, but says nothing about retention or training use.
- What are the best AutoGPT alternatives?
- The closest alternatives we have reviewed are n8n, Dify, Langflow and CrewAI.
Answers checked Nov 13, 2026.