Skip to main content

​
Install OpenManus

Set up the OpenManus codebase and install its dependencies first. If you can already run OpenManus locally, continue to the configuration step.

​
Prerequisites

  • OpenManus is installed and can start locally
  • You have an active AllTokens API Key
  • The target model supports function calling
  • If you plan to use vision tasks, make sure a vision model is enabled too

​
Configure OpenManus

Open the config/config.toml file and point the model settings to AllTokens. Start with a function-calling-capable general model such as Qwen/QwQ-32B, then switch to a vision model only if you need it. 
[llm]
model = "Qwen/QwQ-32B"
base_url = "https://alltoken.co/v1"
api_key = "YOUR_API_KEY"
max_tokens = 16384
temperature = 0.6

[llm.vision]
model = "Qwen/Qwen2-VL-72B-Instruct"
base_url = "https://alltoken.co/v1"
api_key = "YOUR_API_KEY"

​
Run OpenManus

Once the config is in place, start the main program: 
python main.py
OpenManus will break down the task, search for information, extract content, and call tools when needed.

​
Typical workflow

OpenManus works best for multi-step tasks such as research, analysis, and report generation. A typical run usually includes:
  1. Understanding the request
  2. Searching for sources
  3. Opening pages or documents
  4. Cleaning up data with code
  5. Producing the final output
If you need web access, make sure your model supports tool calling and your search tools are available.

​
Troubleshooting

​
The agent does not call tools

Check whether the selected model supports function calling and confirm base_url is set to https://alltoken.co/v1.

​
The model cannot be found

Make sure the model name in config.toml matches the name enabled in your AllTokens account.

​
Vision tasks fail

Verify that the model under [llm.vision] is enabled and supports image input.

​
It crashes immediately after startup

Recheck your dependencies and make sure the API key is valid.

​
Tips

  • Test a simple task first, then move on to more complex reports
  • Separate text and vision testing so failures are easier to debug
  • For production use, give OpenManus its own API Key
  • Choose a more stable general-purpose model if you switch models often

​
Summary

OpenManus is a good fit for multi-step, tool-driven agent tasks. With AllTokens, you can keep model access behind one gateway and choose the right model for each job.