OpenCopilot
OpenCopilot is an AI copilot tool designed for SaaS products. It enables users to build their own AI copilot that can communicate with unde...
Last verified:
What is OpenCopilot?
OpenCopilot is a free and open-source AI copilot tool that allows SaaS product owners to create custom AI assistants tailored to their specific product. The copilot integrates with underlying APIs and can execute API calls whenever needed, using Large Language Models (LLMs) to determine if a user's request requires calling an API endpoint, which endpoint to call, and what payload to pass based on the given API definition.
Key features include Swagger OpenAPI 3.0 support for bulk API import, automatic payload field population based on context, conversation memory to remember context, copilot history to view user interactions, LangSmith traceability for seeing the copilot's thinking process, critical action confirmation before deleting data, complex flow execution calling multiple API endpoints simultaneously, and a user-friendly chat bubble that can be integrated into SaaS apps. The tool also supports flows definition for intricate system processes like
OpenCopilot pricing
Pricing model: Paid
OpenCopilot is free and open-source under MIT license. The tool itself is 100% free for the open-source version. There is also a managed cloud version at cloud.opencopilot.so with a freemium model including free tier and free trial options. The open-source version can be accessed locally at localhost:8888 after installation.
OpenCopilot pros
- Free and open-source under MIT license
- Integrates with underlying APIs to execute calls
- Uses LLMs to determine which API endpoint to call
- Supports Swagger OpenAPI 3.0 for bulk import
- Automatically populates request payload fields based on context
- Transforms API responses into meaningful text
- User-friendly chat bubble for SaaS integration
- Less than 10 lines of code for web/desktop setup
- Copilot memory remembers conversation context
- Copilot history shows user interactions
- LangSmith traceability for fine-tuning prompts
- Asks confirmation before critical actions like deleting data
- Executes complex flows calling multiple endpoints simultaneously
- Flows editor for defining custom system processes
- Supports both GPTs and open-source LLMs
- REST API with streaming support
- Knowledge base integration for training AI
- Monitoring and evaluation tools built in
- Debugging tools for detailed insight
- Python code interface for defining copilots
OpenCopilot cons
- Not suitable for handling large APIs without JSON transformers
- Cannot call multiple endpoints simultaneously in current version
- Does not retain chat history in older versions
- Treats each message as standalone interaction in some versions
- Swagger files treated as second class citizen vs UI
- No UI flows builder yet (planned feature)
- External data sources feature still coming
- Dynamic context feature coming soon not yet available
Frequently asked questions about OpenCopilot
What is OpenCopilot?
OpenCopilot is a free and open-source tool that allows you to create AI copilots for your SaaS product. The copilot can communicate with your underlying APIs and execute calls as needed. It uses LLMs to determine if a user's request requires calling an API endpoint, which endpoint to call, and passes the appropriate payload based on the given API definition.
Is OpenCopilot really free?
Yes, OpenCopilot is free and open-source under the MIT license. You can clone the Git repository and run it locally at localhost:8888. There is also a managed cloud version at cloud.opencopilot.so with freemium options including a free tier and free trial.
How do I integrate OpenCopilot with my SaaS product?
You need to provide your APIs/actions definition including public endpoints and how to call them. OpenCopilot supports Swagger OpenAPI 3.0 for bulk import. After validating your schema, you can integrate the user-friendly chat bubble into your SaaS app with less than 10 lines of code for web or desktop applications.
What API standards does OpenCopilot support?
OpenCopilot supports Swagger OpenAPI 3.0 for bulk API import and is open-source under MIT license. It also supports OpenAPI standards generally, making it compatible with standard API definition formats.
Can OpenCopilot execute multiple API calls at once?
In the current version, OpenCopilot is not suitable for calling multiple endpoints simultaneously. However, the roadmap includes a 'Complex flows' feature that will enable executing complex flows including calling multiple API endpoints at the same time and passing data between these calls.
How does OpenCopilot handle conversation memory?
OpenCopilot has copilot memory that allows it to remember conversation context. You can also ask for modifications on previous requests. The system manages conversation history for you and includes copilot history features to see how users interact with your copilot.
What happens when a user requests a critical action like deleting data?
OpenCopilot includes a 'Critical actions' feature where the copilot always asks for user confirmation before doing critical actions that involve deleting data, ensuring safety and preventing accidental data loss.
Can I track how my copilot thinks and make improvements?
Yes, all copilots are equipped with LangSmith traceability, which allows you to see the thinking process of the copilot and fine-tune your base prompts to enhance results. The tool also includes evaluation methods to measure improvements and debugging tools for detailed insight.
What LLMs can I use with OpenCopilot?
OpenCopilot supports LLM selection from both GPTs (like GPT-3.5-turbo-16k and GPT-4) and open-source LLMs (OSS LLMs). You can specify which LLM to use when creating your copilot in Python code.
How do I define custom flows for my copilot?
After uploading your Swagger file, you can use the flows editor to specify intricate system flows like 'add to cart' or custom processes. You start with an empty flows area, see all your endpoints after uploading Swagger, then create your own flow by putting your API endpoints in sequence. Users can then ask questions like 'Please order me pets with tag pending'.