A visual builder from the Japanese startup ROUTE06 that lets you create and run AI agents without writing code. By dragging and dropping nodes on a canvas and connecting them, you can assemble agents that automate a chain of work such as research, PRD writing, code review, and documentation updates. A key trait is that multiple models — GPT, Claude, Gemini and others — can be used side by side within a single workflow, and with GitHub integration an agent can be triggered by an issue or a pull request. It was officially released in December 2025 as open source under the Apache License 2.0, and is offered both as a self-hosted edition you run on your own server and as a ready-to-use cloud edition.
Key Features
- Visual agent builder: A drag-and-drop editor where you place nodes and link them with lines, so you can lay out an agent’s flow of work without writing code. The flow you build can be run and deployed as is
- Combining multiple models: Supports several foundation models including GPT, Claude, and Gemini, letting you pick the model that suits each task and mix them inside one workflow. You are not locked to a single model vendor
- GitHub AI Operations: Agents can be embedded into the development flow around issues, pull requests, and deployments. After letting an agent read the repository, you can automate review comments or documentation updates
- Ready-made built-in apps: Deep Researcher for competitive and product research, Code Reviewer for automatic PR analysis, PRD Generator for drafting specifications from GitHub activity, and Doc Updater for keeping documentation in step with code changes
- Knowledge store (vector stores / data stores): Internal documents and GitHub repositories can be ingested for agents to reference, making it easier to produce answers and deliverables grounded in your own information
- Two tracks: open source and cloud: The core is published under the Apache License 2.0, so you can run it without limits in your own environment, or use the managed cloud edition instead
Pricing
| Plan | Monthly price | What’s included |
|---|---|---|
| Open Source (self-hosted) | Free | Unlimited seats, vector stores, and data stores. You provide the server and the API keys for each model |
| Free (cloud) | $0 | 1 user, up to 30 minutes of AI model usage per month, basic models only, community support |
| Pro | $20 | 1 user, $20 of monthly usage credits, 5 document vector stores / 3 GitHub vector stores / 10 data stores, premium models, pay-as-you-go overages, email support |
| Team | $100 (coming soon) | Up to 10 users, 20 document vector stores / 10 GitHub vector stores / 20 data stores, priority email support |
Pricing is current as of August 2026. Please check the official site for the latest information.
Pros & Cons
✅ Pros
- You can assemble multi-step AI agents on screen without writing code
- Being open source under the Apache License 2.0, it can be run on your own servers free of charge and without limits
- GPT, Claude, and Gemini can be used selectively within one workflow, which also makes switching models easier
- GitHub integration is built in from the start, so it slots directly into a development team’s daily work
- Finished examples such as Deep Researcher and Code Reviewer are bundled and ready to try
⚠️ Cons
- The cloud free tier is only 30 minutes a month, so real-world use will require a paid plan fairly quickly
- The Team plan is still listed as coming soon, so options for multi-person use are limited at present
- The self-hosted edition is free, but you bear the cost of running the server and of each model’s API usage
- Its GitHub-centric design can make it a less natural fit for work outside software development
- It is a relatively new service, and Japanese-language articles and case studies are still few
Comparison with Similar Services
| Item | Giselle | Dify | n8n | Flowise |
|---|---|---|---|---|
| Provider | ROUTE06 (Japan) | LangGenius (China) | n8n GmbH (Germany) | FlowiseAI |
| License | Apache License 2.0 | Open source (with some restrictions) | Sustainable Use License | Apache License 2.0 |
| Main use | Building AI agents, automating GitHub work | Building LLM apps and chatbots | General-purpose workflow automation | Visual building of LLM apps |
| GitHub integration | Built in as a core feature | Available (general-purpose) | Available (general-purpose) | Available (general-purpose) |
| Cloud free tier | Yes (30 min/month) | Yes | Yes | Mainly self-hosted |
Who Is It For
- Product teams and planners who want to prototype AI agents without writing code
- Development teams that want to hand review and documentation updates on GitHub over to AI
- Anyone who wants to build in multiple models for comparison rather than depend on one model vendor
- Companies that need to self-host because they handle internal data
- People who value a Japan-based tool with information and support available in Japanese
Summary
Giselle is a visual builder where you assemble AI agents on a drag-and-drop canvas and take them straight through to running and deployment. Its strengths lie in mixing multiple models, its GitHub integration, and being open source under the Apache License 2.0. A good approach is to first try the built-in apps on the cloud free plan, then choose between self-hosting and the Pro plan once you move to serious use.