Evaluate the Platform; keep Classic for study
AutoGPT Classic is unsupported, and its dependencies will not be updated. Its README at the tested commit recommends educational use and directs prospective users to the Platform. That maintenance warning is sufficient reason to avoid choosing Classic for a new deployment, regardless of whether installation completes.
The current AutoGPT README describes specialists whose instructions, context, skills and connected apps can be configured, with Otto coordinating work. AutoGPT's visual builder lets you connect blocks, branch logic and control workflow steps. These are documented capabilities, not outcomes demonstrated by the sandbox.
What the September 28 run establishes
Our September 28, 2026, Classic run at c035ab1 recorded pytest exiting 1. The matching sandbox record identifies the classic subdirectory, 3 CPUs, 8 GB of memory and a Python 3.12 container image; the install tail identifies Python 3.12.12.
Installation returned zero in 80 seconds. The build step returned zero in 7 seconds, but its retained tail is empty. Neither result demonstrates that an autonomous agent, CLI workload or Platform deployment completed useful work. The installed environment contained 175 packages and occupied 786 MB.
Pytest exited 1 after 56 seconds according to the harness. Its stored parser summaries report 277 passed, 65 failed, 4 skipped, 48 errors and a total field of 390. These are recorded parser summaries, not verified exhaustive counts of unique tests. Skipped tests are separate from failures.
The raw footer also lists 14 xfailed, 277 warnings and 42.38 seconds. These remain separate from the harness values; they are not combined into a new total or pass rate. No new execution or complete-log reconciliation was performed.
The retained log names local file-storage rename and copy tests, browser-component tests, and original AutoGPT configuration and permission-feedback tests. Their repeated failure message is "async def functions are not natively supported." The evidence does not establish why async support was unavailable or whether those tests would pass in another environment.
The publisher's testing methodology explains command recording and sandbox limits. It supplies provenance context, but does not establish that its current procedure was used unchanged for this older run.
Read the audit as an advisory record
Pip-audit recorded 108 advisories in the Classic environment. The record does not include scannedPackages, so scan coverage is unknown. Its packages field of 5 must not be treated as scan coverage.
The result supports investigating the installed dependencies, alongside Classic's explicit warning about known dependency vulnerabilities. It does not establish exploitability, identify a new security finding or assess the Platform's security. Classic's documented workspace permissions and interactive approvals do not resolve those dependency questions.
Match setup instructions to the checkout
The current README advertises the significantgravitas/autogpt single-container image, followed by account creation and administrator promotion. It also describes installers for Linux, macOS and Windows. Self-hosters provide model credentials and manage updates and operation; the hosted Platform supplies infrastructure and model access as a paid service.
By contrast, the guide pinned to c035ab1 says the appliance endpoint and image tags were not public. Its manual path requires Git, Node.js, NPM, Docker and Docker Compose. It documents make init-env for generating installation secrets, plus Postgres, Redis and RabbitMQ services.
The older release warning describes the pinned checkout, not the current installer's status. Neither Platform installation path was tested in the recorded run. For a Platform trial, follow the instructions matching the selected release, then verify a representative workflow, credential access, recovery and upgrades before relying on it.
Check licensing before embedding the Platform
The current repository's license table assigns Polyform Shield to autogpt_platform/ and MIT to classic/ and everything else. It describes Platform use as free for personal and internal business work, with a restriction against selling a competing hosted service. Check the license against the intended distribution and service model before building around it.
Alternatives for retrieval and application integration
Dify combines visual AI workflows with a RAG pipeline. Its README describes document ingestion and retrieval, making it a relevant alternative when retrieval is central to the application. This evidence does not establish that Dify is faster, safer or easier to operate than AutoGPT.
Langflow supports visual authoring and exposes workflows through API and MCP servers. Its README makes it relevant when a flow needs to become a tool inside another application. Evaluate it for that integration need; the supplied evidence does not establish equivalence with AutoGPT's specialist experience.
Neither alternative was tested in this AutoGPT run.
What the ratings assess
The ratings are editorial judgments within the scopes below. They are not a Platform benchmark or a security assessment. Setup uses installation and verification outcomes; documentation uses the supplied explanations of status, setup and licensing; community uses documented attention and participation routes. On the five-point scale, 1 indicates severe obstacles or little support within the assessed scope, 3 indicates mixed evidence, and 5 indicates strong support. Intermediate scores reflect the balance described in each note. An unassessed dimension receives no number.
Setup ease is 2/5 for the tested Classic environment: installation completed, but pytest exited 1. This does not rate current Platform setup.
Documentation is 4/5 because the supplied sources explain product status, setup and licensing. Readers still need to distinguish current instructions from those pinned to the tested commit.
Community is 5/5 for the narrow dimension of public attention and available participation routes. The README pinned to c035ab1 advertises 185,000+ GitHub stars and lists Discord, issue reporting and contribution routes. This rating does not measure adoption, response times or support quality.
Maturity is unassessed. The recorded run covered Classic, which is unsupported, and did not validate a Platform workload. The supplied primary-source extracts do not establish the claimed Platform release version or beta status, so those assertions cannot support a numerical maturity rating.