What it does
7 features
AutoGPT · Platform for AI agents and workflows
Product and access
AutoGPT consists of the platform (Polyform Shield) and the older AutoGPT Classic (MIT), which the documentation says is no longer supported. Determinist Ltd, trading as AutoGPT, is responsible for the hosted platform.
Core features
7 features from provider sources · checked on 5 November 2026. Availability and limits are directly below.
AutoPilot and Otto
AutoPilot carries out supported actions in conversation: running agents, building from a description, searching the marketplace and setting up schedules. According to the website, Otto coordinates specialist Experts.
Source 6Schedules and webhook triggers
Agents run at set times or start when an external service sends data to a webhook URL.
Source 7Marketplace and library
Ready-made agents can be added from the marketplace to your own library and run from there.
Source 10API and MCP server
Through an API and an MCP server, AI tools can search, build and run agents and read results. Access uses an API key with permissions.
Source 12Human review within a run
A block can pause a run until a person approves or rejects data. According to the documentation, the run waits indefinitely.
Source 11
Show all 7 features
Self-hosting with Docker
A single container holds the interface, backend and databases; state is kept in a Docker volume.
Source 4
Usable AI models
Open choice
In the cloud, models are built in, the Pro plan promises access to virtually any leading model according to the provider, and you pick them in the AI blocks. Custom endpoints and local models work only when self-hosting: there you use your own keys, Ollama for blocks and any OpenAI-compatible endpoint for chat. Ollama models outside the default list need their own catalog entry.
Model families named
- Anthropic (Claude)
- OpenAI (GPT)
- Google (Gemini)
- xAI (Grok)
- Meta (Llama, Muse)
- Mistral
- Cohere
- Microsoft (MAI)
- Z.ai (GLM)
- DeepSeek
- Alibaba (Qwen)
- Moonshot (Kimi)
- Other models can be connected freely
- Model choice
- You choose
- Your own API key
- Yes
- Router or custom endpoint
- Yes
- Local models
- Yes
- Depends on
- Plan
The rating covers the generally available main access. Documented · Sources accessed on
AutoGPT: Pricing · AutoGPT: AutoPilot with a local LLM · AutoGPT: Using Ollama · AutoGPT: Built-in blocks · AutoGPT: README
Does it fit my task?
2 practical tasks · 7 requirements and limits
A good fit if …
- you want to build recurring processes as visual workflows with a schedule or webhook trigger
- you want to run the platform on your own infrastructure with your own model keys
- you want to control agents from an AI tool through the MCP server
Less suitable if …
- you expect free cloud use without a card on file
- you must process customer data in a documented EU hosting region
- you need ready multi-user and admin management today
Editorial assessment based on provider documentation. Not a product test.
Practical tasks
Editorial examples based on documented features. No product tests performed.
Recurring research on a schedule
- Your input
- An agent with input and output blocks and fixed inputs for the schedule.
- Possible result
- The agent runs at the set times in your timezone.
- What to check
- Read the run results and check that all required outputs are present.
Webhook starts agent with approval
- Your input
- An agent with a trigger block and a block for human approval.
- Possible result
- The run starts on incoming data and pauses before the approval.
- What to check
- Answer waiting reviews; without an answer the run waits indefinitely.
Build a recurring research task as an agent and run it on a schedule.
Your possible prompt
Design a process for new requests: capture information, sort it by topic, and create a proposed response. Plan an approval step before external actions.
Adjust the task, data and permissions.
Requirements and limits
Requirement
Paid cloud service
The hosted platform is a paid service. Eligible new accounts can start Pro with a 7-day trial that requires a card; workflow runs draw on credits.
Source 2Requirement
Self-hosting: Docker and model keys
You need Docker, according to the guide more than 5 to 6 GiB of RAM, and model access of your own for AutoPilot and memory: OpenRouter plus OpenAI, an Anthropic key in addition to both, or a local OpenAI-compatible server.
Source 4Limitation
Single container is not multi-tenant
The guide says explicitly that the container is not a turnkey multi-tenant distribution. Uploaded files are not scanned for malware, and you set up TLS for remote access yourself.
Source 4Limitation
Managed services missing when self-hosting
According to the pricing page, hosted billing, managed credentials and marketplace publishing are not part of self-hosting.
Source 2Limitation
AutoGPT Classic no longer supported
According to the documentation, Classic is not supported from a security perspective; dependencies are not updated and issues are not fixed.
Source 14Limitation
Completed run does not mean error-free
A run can count as completed even though a block failed. You check that the outputs you need are present.
Source 11Not yet confirmed
Team features
The pricing page lists Team as coming soon, with multi-user workspaces and admin roles; contact is through sales.
Source 2
Can I use it?
Availability still unclear
Europe & regional availability
Privacy policy, subprocessor list, terms of use and pricing page reviewed. No statement on availability in Europe was found.
Reviewed scope: Hosted platform and self-hosted installation
Reviewed on:
Availability still unclear
Availability in Europe: no statement
The reviewed pages name no country or region release for sign-up and payment. The terms are governed by the law of England and Wales; mandatory consumer protections of your country of residence are preserved according to the terms.
Region, reason and source
Region covered: EU/EEA, United Kingdom and Switzerland
Documented reason: The reviewed sources name no country release for this.
Availability still unclear
Data processing mainly in the US
According to the list, Google Cloud, Supabase and the model providers OpenAI and Anthropic are in the US. The list names no EU hosting option for the platform. Langfuse (AutoPilot traces) and a PostHog instance are among those located in the EU. For transfers the page cites standard contractual clauses or the EU-US Data Privacy Framework; the privacy policy refers to the UK GDPR.
Region, reason and source
Region covered: EU/EEA
Documented reason: The sources make no statement on a hosting region in Europe for customer data.
Availability still unclear
Self-hosting: data stays with you
When self-hosting, all durable state lives in your own Docker volume. The reviewed pages say nothing about Europe; you choose model providers and their region yourself.
Region, reason and source
Region covered: EU/EEA
Documented reason: The reviewed sources name no country release for this.
Accounts & permissions
Based on provider documentation. Permissions depend on the account, integration and organization policies. Review the permissions actually requested before use.
Configuration covered: Cloud account on the platform or your own installation
Sources reviewed:
Show all 3 points
Documented
Data access & team sharing
Agents use the connections you give them. The MCP server acts on your AutoGPT account and lists only the tools the API key’s permissions allow; some tools need no permission. The key reaches only its own organisation’s resources. When connecting MCP services, the page advises checking the requested permissions.
Evidence · Evidence
Memory and conversation continuity
0 of 6 criteria documented
When self-hosted, AutoPilot memory sits in a graph database in the Docker volume; inspection and deletion are documented only for learned skills.
AutoPilot memory and learned skills of the platform
Sources accessed:
Documentation by criterion
StoragePartly documented
Storage and scope
When self-hosted, FalkorDB stores memory under /data/falkordb in the Docker volume. Learned skills are saved as immutable versions, at most 150 per account or Expert. The reviewed pages do not say where the cloud platform stores memory.
SelectionPartly documented
Selection and timing
Nightly skill learning is off by default and enabled per cohort through a feature flag. It runs at 03:00 in the person’s timezone and proposes skills from verified work. The pages do not describe how AutoPilot memory selects what to keep.
Edit & deletePartly documented
Inspect, edit and delete
For skills you can view history and sources, pause learning, edit, restore a version and delete; individual conversations can be excluded from learning. The pages describe no controls for individual memory entries.
Access boundariesPartly documented
Agent and channel boundaries
Skills are private to the account and the Expert that produced them; Experts keep separate conversations. Personal chats teach the personal assistant, Expert chats only that Expert. Separation of graph memory per person is not described.
RetrievalNot publicly established
Retrieval
The reviewed pages do not describe how memory is retrieved when answering. Extracting memory needs a model and an embedding model, for example a local one such as nomic-embed-text.
New conversationsPartly documented
Session reset and compaction
Resetting memory per person is not documented. When self-hosted, all durable state lives in the volume under /data, and each installation should use its own volume. The pages say nothing about compaction of long chats.
Cost and effort
5 plans · Prices as of 5 November 2026
Prices and included usage
Pricing checked
Cloud: Pro $50 per month after a 7-day trial, Max $320 per month, Team announced. Workflow runs are paid from a credit wallet on every plan. Self-hosting is free.
| Plan | Price | Included |
|---|---|---|
| Pro | $50 per month, after a 7-day trial for eligible new accounts | Access to models, Experts, Otto chat, visual builder, connections and scheduled workflows; chat usage is covered by the subscription. Price source |
| Pro (annual) | $510 per year, equal to $42.50 per month | As Pro, billed annually. Price source |
| Max | $320 per month | Everything in Pro, plus 8.5x Pro usage, 5x file storage, early feature access and priority support. Price source |
| Team | Coming soon, contact sales | Multi-user workspaces, admin roles, centralised billing and approvals are listed. Price source |
| Self-hosted | No software fee stated | Platform via Docker with your own infrastructure and model keys. Price source |
- The pricing page does not clearly show which plans can top up credits.
Setup
Create an account on the platform or run the Docker container yourself and add your own model keys.
Operation & oversight
Based on public provider documentation. Assessments are editorial inferences from the stated properties. Actual time requirements and reliability have not been measured.
Deployment covered: Cloud operation and self-hosting
Sources reviewed:
Editorial assessment
Schedules, triggers and reviews are documented. When self-hosting you are responsible for backups, TLS and updates, and failure notifications are not described.
Show all 6 points
Documented with conditions
Sign-ins
You connect services by OAuth, API key or password. The reviewed pages do not say how long access lasts or how it is renewed.
EvidenceDocumented
Human intervention
A block can put a run into REVIEW until a person approves or rejects; the run waits indefinitely. If your integration cannot answer, the documentation names handing over to a person, answering yourself or stopping the run. For unattended agents, reviews can be turned off, and then every review counts as approved.
EvidenceDocumented with conditions
Resuming after failures
When self-hosting, you take a cold backup before an upgrade, replace the container with the same volume and, if needed, restore the matching archive into a new volume. You must not run an older image against an already migrated volume. Runs still executing when the container stops are abandoned and do not resume; you start a new run.
Evidence
Evidence and open questions
18 sources · 4 open points
Open points
The reviewed sources contain no reliable information on these points. A gap says nothing about the product.
Europe: Availability still unclear
- Availability in Europe: no statement
- Data processing mainly in the US
- Self-hosting: data stays with you
Requirements and limits: not yet confirmed
Sources and evidence
Official website, pricing page, documentation, legal pages and repository reviewed. Feature descriptions come from the provider; example tasks are editorial. Hands-on tests are still pending.
Editorial team: Michael Meierhoff · sources reviewed on . Provider information is not independent evidence of performance.
Provider
Determinist Ltd offers the platform under the name AutoGPT; the source code is in the Significant-Gravitas/AutoGPT repository.
- Operator
- Determinist Ltd (AutoGPT)
- Organization
- Provider
- Website
- agpt.co
- Repository
- Significant-Gravitas/AutoGPT
- GitHub stars
- 187.909 · Reviewed
- License
- Polyform Shield (autogpt_platform), MIT (classic and rest)
Changes to this profile 1
A platform for building and running your own agents, hosted or self-hosted.
These notes describe editorial changes. Source and pricing review dates appear in their respective sections.