What it does
7 features
OpenManus · Open-source agent framework
Product and access
OpenManus is a community project of the GitHub organisation FoundationAgents. The README calls it a simple implementation and says Manus can be tried without an invite code; core authors are from MetaGPT, according to the README. The latest commit on the main branch dates from 16 August 2026 (GitHub API, checked on 5 November 2026). The repository is not archived, but development is quiet: the latest release, v0.3.0, is from 10 April 2025.
Core features
7 features from provider sources · checked on 5 November 2026. Availability and limits are directly below.
General-purpose agent in the terminal
You start the agent with python main.py and enter your idea in the terminal.
Source 1Browser automation through Browser Use
OpenManus starts Browser Use CLI 3.0 as an MCP server. Locally it attaches to Chrome or Chromium without an API key; a cloud browser needs a Browser Use key. OPENMANUS_DISABLE_BROWSER_USE=1 turns it off.
Source 1Multiple agents in flow mode
python run_flow.py starts the multi-agent version, which the README calls unstable. A data analysis agent is off by default and can be enabled in config.toml.
Source 1Configurable models
The example configuration contains entries for Anthropic, Amazon Bedrock, Azure OpenAI and Ollama, among others, plus an optional vision model.
Source 3Optional Docker sandbox
The example configuration has a commented-out [sandbox] section with Docker image, memory and CPU limits, timeout and network switch; use_sandbox is set to false there.
Source 3
Show all 7 features
Selectable search engine
According to the example configuration, the search engine can be set to Google, Baidu, DuckDuckGo or Bing, with a fallback order.
Source 3
Usable AI models
Open choice
You enter the model, base URL and key in the configuration yourself. The example file shows Anthropic, OpenAI via Azure, Bedrock, the Jiekou.AI gateway and Llama locally via Ollama. No model access is included, so you need your own account or a local server.
Model families named
- Anthropic (Claude)
- OpenAI (GPT)
- Meta (Llama, Muse)
- Other models can be connected freely
- Model choice
- You choose
- Your own API key
- Yes
- Router or custom endpoint
- Yes
- Local models
- Yes
The rating covers the generally available main access. Documented · Sources accessed on
Does it fit my task?
2 practical tasks · 5 requirements and limits
A good fit if …
- you want to run an open-source general-purpose agent yourself and try it with your own model key
- you want to plug Browser Use, MCP or your own tools into a Python agent
- you want to read and adapt the code and can live with a quiet pace of development
Less suitable if …
- you expect a hosted product with an account, support and administration
- you need documented memory, schedules or approval per action
- you need stable multi-agent operation without experiments
Editorial assessment based on provider documentation. Not a product test.
Practical tasks
Editorial examples based on documented features. No product tests performed.
Research idea in the terminal
- Your input
- A question or idea you type into the terminal after python main.py.
- Possible result
- The agent works through the task with the configured model and its tools.
- What to check
- Check sources, page visits and results yourself, especially with the browser tool active.
Data analysis in flow mode
- Your input
- A file or task to analyse, data analysis agent enabled in config.toml.
- Possible result
- Analysis and charts through run_flow.py, which the README calls unstable.
- What to check
- Compare numbers and charts with the raw data and install the extra dependencies.
Install OpenManus locally and try a research idea in the terminal.
Your possible prompt
Compare three providers for [task]. Use their official documentation. Show differences, sources, and open questions in a table.
Adjust the task, data and permissions.
Requirements and limits
Requirement
Python 3.12 and your own model key
Both install routes in the README require Python 3.12. Without an entry in config.toml the agent has no model.
Source 1Limitation
Key in the configuration file
The README has you enter the API key directly in config/config.toml.
Source 1Limitation
Sandbox not active by default
In the example configuration the sandbox section is commented out and use_sandbox is false.
Source 3Not yet confirmed
Privacy, schedules and memory
The README describes no privacy statement, no schedules and no memory across sessions.
Source 1
Can I use it?
Availability still unclear
Europe & regional availability
README, example configuration and project page reviewed. They name no country release or data region.
Reviewed scope: Local software; model providers separately
Reviewed on:
Availability still unclear
Europe: no statement from the project
OpenManus runs on your own machine. The reviewed pages say nothing about Europe. What data goes to model providers or to the optional Browser Use cloud browser depends on your configuration; the README does not describe this.
Region, reason and source
Region covered: EU/EEA, United Kingdom and Switzerland
Documented reason: The reviewed sources name no country release for this.
Official source · github.com · Official source · raw.githubusercontent.com
Accounts & permissions
Based on provider documentation. Permissions depend on the account, integration and organization policies. Review the permissions actually requested before use.
Configuration covered: Self-run agent with your own model key
Sources reviewed:
Show all 3 points
Documented
Account used
You enter the model key directly in config/config.toml. A cloud browser additionally needs a Browser Use key as an environment variable.
Evidence
Memory and conversation continuity
0 of 6 criteria documented
The reviewed pages describe no memory and no stored sessions.
Self-run agent framework
Sources accessed:
Documentation by criterion
StorageNot publicly established
Storage and scope
The README describes no storage location for memory or sessions. The example configuration mentions only memory limits for the optional sandbox.
SelectionNot publicly established
Edit & deleteNot publicly established
Access boundariesNot publicly established
Agent and channel boundaries
Not described. The README names the launch variants main.py, run_mcp.py and run_flow.py, but no memory boundaries between them.
RetrievalNot publicly established
New conversationsNot publicly established
Session reset and compaction
Not described; the README names neither session end nor compaction.
Cost and effort
1 plan · Prices as of 5 November 2026
Prices and included usage
Pricing checked
OpenManus is under the MIT licence; the reviewed pages name no price. Costs arise for model access with the chosen provider.
| Plan | Price | Included |
|---|---|---|
| OpenManus | No software fee stated | Agent framework with your own model access. |
- The optional Browser Use cloud browser needs its own key according to the README; terms were not reviewed.
Setup
Clone the repository, install dependencies with conda or uv, create config.toml from the example file and enter a model key.
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: Local operation in the terminal
Sources reviewed:
Editorial assessment
You run and supervise OpenManus yourself. The reviewed pages describe no approvals, memory or error handling; plan your own isolation, for example through the optional sandbox.
Show all 6 points
Open in reviewed sources
Files and working state
The README does not describe whether or where session state is kept.
EvidenceDocumented
Starting further runs
You start runs yourself with python main.py, run_mcp.py or run_flow.py. The README describes no schedules or triggers.
EvidenceOpen in reviewed sources
Human intervention
The README describes no confirmation before actions. Input happens in the terminal.
EvidenceOpen in reviewed sources
Resuming after failures
The README does not describe how to continue interrupted work.
Evidence
Evidence and open questions
4 sources · 2 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
Requirements and limits: not yet confirmed
Sources and evidence
README, example configuration, licence and project page reviewed. Commit date, release and stars come from the GitHub API. 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
Open-source community project on GitHub under the MIT licence. The project page names no operators; the README points to a contact within the project.
- Operator
- FoundationAgents
- Organization
- Provider
- Website
- openmanus.github.io
- Repository
- FoundationAgents/OpenManus
- GitHub stars
- 58.891 · Reviewed
- License
- MIT
Changes to this profile 1
Open-source framework for a general-purpose agent that you install yourself and start in the terminal with your own model key.
These notes describe editorial changes. Source and pricing review dates appear in their respective sections.