Giselle
Giselle is a no-code AI app builder from Tokyo-based ROUTE06: drag-and-drop nodes chain Anthropic, Google and OpenAI models into agents that read your own data and automate GitHub work. Open source, with a permanent free plan.
What is Giselle?
Giselle is an AI app builder published by ROUTE06, Inc., a Japanese company founded on 24 January 2020. Its documentation describes it as a platform that lets anyone build AI applications intuitively, through drag-and-drop, without any programming knowledge; the terms of service call it a generative AI service for the no-code development of agents and workflows. A second positioning runs alongside the first: Giselle presents itself as an execution platform designed to complete complex, long-running AI tasks reliably, as business operations rather than one-off prompts.
Work happens in a visual node editor. You place nodes on a canvas and wire them like a flowchart, add a Generator Node and choose its model in the properties panel. Five core capabilities are documented: Visual Node Editor, Long-Running Task Execution, Multi-Model AI Collaboration, GitHub Integration and Custom Knowledge. The marketing site covers the same ground in four product blocks, Multi-Model Composition, Visual Agent Builder, Knowledge Store and GitHub AI Operations, and packages four ready-made scenarios: Deep Researcher, PRD Generator, Code Reviewer and Doc Updater.
The multi-model angle is the distinctive part. More than fifteen models from four providers, Anthropic, Google, OpenAI and fal.ai, can be combined inside a single workflow, each given a role, with automatic model selection, cost-oriented routing and performance monitoring; fal.ai models are reserved for the open-source edition. Knowledge is handled without RAG expertise: Document Vector Stores, GitHub Vector Stores and PostgreSQL Data Stores are indexed, embedded and searched semantically, with automatic synchronisation.
GitHub is the second differentiator. Workflows fire on issues, pull requests and comments, and repository commands such as /code-review and /prepare-release run against the codebase, which can itself be indexed as knowledge.
The audiences named by the publisher are broad: engineers, designers, product managers, marketers, sales and back-office professionals. Two facts temper the picture. The Free and Pro plans are single-user, with the Team plan still Coming Soon, and the API and its SDK are in Private Preview. The core is open source, self-hostable from the giselles-ai/giselle repository with unlimited quotas, and the publisher holds ISO/IEC 27001 certification under certificate IS 779274.
What it does
- Build AI agents by dragging and connecting nodes, with no code and no prompt tuning
- Chain several agents together into multi-agent workflows
- Choose the model for each task, or let Giselle route automatically across Anthropic, Google and OpenAI
- Feed agents your own data: PDFs, documents, GitHub code, issues and pull requests, PostgreSQL databases
- Trigger workflows from GitHub events and repository commands
- Run long tasks in a dedicated execution environment with real-time tracking
- Call your apps from outside through the API, or self-host the open-source edition
When to use Giselle / When not to
A quick filter to help you decide if Giselle is the right fit.
When to use Giselle
- AI-native startups and lean product teams that want working AI agents without hiring a machine learning engineer
- Solopreneurs and fast builders, including non-developers such as designers, product managers, marketers, sales and back-office staff, who want to automate repetitive work without writing code
- Product-led engineering teams already living inside GitHub, where issues, pull requests and commands such as /code-review or /prepare-release can trigger a workflow
- Technical writers and DevRel teams keeping documentation, PRDs and release notes in step with the codebase
- Innovation and research teams running long, monitored AI tasks such as market analysis, customer research, technical due diligence or academic research
When not to use Giselle
- Teams that need to collaborate today: the Free and Pro plans are capped at one user each, and the Team plan for up to 10 users is still announced as Coming Soon
- Organisations that expect phone or live chat support, since Giselle answers by email only
- Buyers who need a stable, generally available API or a mobile app: the API is in Private Preview with access limited to selected users and may change without notice, and there is no iOS or Android application
- Companies bound by data residency requirements, as the privacy policy allows information to be transferred, processed and stored anywhere in the world
- Anyone planning to resell generated Output as a standalone product, to build models competing with ROUTE06, or to submit sensitive data such as identification numbers, banking credentials, biometric, health or criminal record data, all of which the terms prohibit; users under 18 are also excluded without parental consent
How to use Giselle
A typical end-to-end flow, from setup to results.
- Create an account on the Giselle studio with an email address, or with a Google, Microsoft or GitHub login
- Follow the documented quickstart, which runs from sign-up to a first AI app actually executed
- Open a Workspace and lay out your nodes on the canvas, connecting them like a flowchart
- Add a Generator Node and pick its model in the Properties Panel, then enable Web Search from the node's Tools tab if the task needs it
- Or take the three-step path shown on the Visual Agent Builder page: start from a template or from scratch, build, then deploy in one click
- Connect GitHub under Settings, Team, Integrations with the Configure GitHub App button, granting access to every repository or to a selection, up to 100 repositories per installation
- Add knowledge under Settings, Team: Vector Stores for documents and GitHub repositories, Data Stores for PostgreSQL connections
- Test your agent in the Playground, then use Tasks for monitored runs
- For external use, create a team-scoped API key under Settings, Team, API Keys, in the gsk_ format shown only once, and call POST /api/apps/{appId}/run with an Authorization Bearer header
- To self-host, start from the open-source repository giselles-ai/giselle
Pros & Cons
Pros
- No coding skill required: the entire build takes place in a visual editor
- Four model providers usable side by side in a single workflow, so the catalogue can evolve without rewriting agents
- Open-source core that can be self-hosted with unlimited Vector Store and Data Store quotas, which is a genuine exit route
- Low entry cost: a permanent free plan, then Pro at 20 USD per month including 20 USD of AI credits
- Usage beyond the Pro allowance keeps running rather than stopping, billed at a 10 percent markup on each model's base token price
- Unusually deep GitHub integration for a no-code agent builder, down to commands issued inside repositories
- Readable governance for a young product: publisher certified ISO/IEC 27001 since 12 February 2023, third-party penetration tests, TLS in transit, Output ownership assigned to the user, no claim over Input, a published table of 20 third-party products and documentation in English and Japanese
Cons
- No team collaboration in practice: Free and Pro are limited to one user and the Team plan for up to 10 users is only announced as Coming Soon
- The API and SDK are in Private Preview, with access restricted to selected users and features liable to change without notice
- Support is by email only, with no phone and no chat, and priority support is reserved for the unreleased Team plan
- The opt-out instructions for model training referred to in the privacy policy are published nowhere, and there is neither a data processing agreement nor an Article 27 EU representative
- No hosting country or region is disclosed, and the privacy policy allows information to be transferred and stored anywhere in the world
- AI credits are a monthly allowance that never rolls over, and unused credits are lost when moving from Free to Pro; the free plan stops dead once its 30 minutes of model usage are consumed, and that allowance is presented as promotional and changeable without notice
- Other limits worth knowing: no mobile application, fal.ai models restricted to the open-source edition, SOC 2 announced as on the way rather than obtained, payments non-refundable except where the law requires, and free accounts inactive for more than a year liable to deletion after notice
Pricing & Plans
A free plan is available on a permanent basis, at 0 USD per month, offering up to 30 minutes of AI model usage, base models only and a single user. The lowest paid entry point is the Pro plan at 20 USD per month, which includes 20 USD of AI credits, every premium model and email support. Usage beyond that allowance is not interrupted but billed on the following cycle, at a 10 percent markup on each model's base token price. A Team plan at 100 USD per month is announced as Coming Soon and is not available at present. The open-source edition remains free of charge when self-hosted. Credits do not carry over from one month to the next, downgrades take effect at the end of the billing cycle, payment is collected automatically at each renewal through Stripe, and any price increase requires at least 30 days' notice and applies from the following renewal.
- agent creation and deployment
- multi-agent workflows
- agents enriched with your own data
- no-code builder
- unlimited Vector Stores and Data Stores
- community support through GitHub issues
- 30 minutes of AI model usage
- base models only (Claude Haiku 4.5
- Gemini 2.5 Flash Lite
- GPT-5-nano)
- managed cloud deployment
- 1 user
- no Vector Store and no Data Store
- everything in Free plus 20 USD of AI credits per month
- all premium models (Claude Opus 4.6 and 4.5
- Claude Sonnet 4.6 and 4.5
- the Gemini 3.x and 2.5 range
- GPT-5.4
- GPT-5.3-codex
- GPT-5.2 and 5.2-codex
- GPT-5 and GPT-5-mini)
- everything in Pro plus team collaboration and sharing
- up to 10 users
- 20 Document Vector Stores
- 10 GitHub Vector Stores
- 20 Data Stores and priority email support
Data, GDPR & hosting
A consolidated view of how Giselle handles your data.
GDPR overview
The GDPR is named explicitly. Section 4 of the privacy policy, effective 1 October 2024, allows transfers anywhere in the world while undertaking to safeguard information under applicable laws, including the General Data Protection Regulation. Section 5a grants the eight rights of the regulation: access, erasure, rectification, portability, restriction, withdrawal of consent, objection and complaint to a supervisory authority. They are exercised from the account or by email to support@giselles.ai, free unless a request is manifestly unfounded, repetitive or excessive; EEA, Swiss and UK residents are pointed to their local authority. Section 8 sets out contract performance, legitimate interests, consent and legal obligations as legal bases. Three things are missing: no Article 27 EU representative, no named data protection officer beyond an internal Personal Information Manager, and no data processing agreement anywhere on the site. Japanese law applies, with exclusive jurisdiction in the Tokyo District Court.
Who owns the data?
Section 4b of the terms is explicit: you retain your rights in the Input and own the rights in the Output; ROUTE06 assigns you any rights it may hold and claims none over either. The Output may be used to build commercial tools for yourself or your employer, but it may not be sold or marketed as a standalone commercial product, nor incorporated as a component of commercial products. ROUTE06 keeps the rights to the underlying technology, and section 4c warns that Output is not unique: other users may receive similar results. Feedback you send is licensed to ROUTE06 worldwide, perpetually, irrevocably and royalty-free. Card details are collected and stored by Stripe, Inc.
Reuse rights
Section 4d of the terms lets ROUTE06 use User Content to provide, maintain, develop and improve the Service, comply with the law, enforce the terms and keep the Service secure. Section 2a of the privacy policy adds six purposes: delivering, administering and analysing the Service, improvement and research, communication, new product development, fraud prevention and legal compliance. The legal bases section states that legitimate interests cover developing, improving and promoting the Service, including training our models, drawing on account information, content, social information and technical information, and it refers the reader to instructions for opting out of the use of information for model training. Those instructions appear nowhere on the site, in the documentation or in the FAQ as of 24 August 2026, so the opt-out is announced rather than available. Information is disclosed to vendors and service providers and to analytics services. Section 9c states that ROUTE06 does not sell or share personal information for cross-context behavioural advertising and does not process sensitive data to infer characteristics, and section 1c states that no special categories of data are collected. On the user's side, Output can be reused without asking permission, within the resale limits set by section 4b.
Data retention & training
Hosting summary
No hosting country or region is named anywhere on the site. Section 4 of the privacy policy states that information may be transferred, processed and stored anywhere in the world, in jurisdictions whose data protection laws may differ from the user's, with an undertaking to safeguard it in accordance with applicable laws including the GDPR. What is documented is the infrastructure rather than its location: the Privacy and Security page of the documentation cites secure cloud infrastructure on AWS, Supabase and Vercel, and the table of third-party products in the terms adds Google Cloud, Stripe, Sentry, PostHog, Resend, Langfuse and Trigger.dev among twenty suppliers. The publisher, ROUTE06, Inc., is established in Tokyo, and Japanese law governs the contract. The announced security measures are TLS in transit, role-based access control, rate limiting and third-party penetration tests, backed by ISO/IEC 27001 certification. The domain resolves to 76.76.21.21, an anycast CDN node geolocated in the United States, which indicates where the site is served from and not where data is stored. Organisations with data residency obligations therefore have nothing firm to rely on here.
Things to keep in mind
Risks and trade-offs to weigh before adopting Giselle.
- The Team plan and the prebuilt templates are announced as Coming Soon: check that they exist before building a decision on them
- The privacy policy claims legitimate interests for developing and promoting the Service, including training models, and points to opt-out instructions that are published nowhere; assume your content may be used for training until the publisher confirms otherwise
- No data processing agreement and no Article 27 EU representative, which may be a blocking point for a European controller
- Information may be transferred, processed and stored anywhere in the world, under Japanese law and the exclusive jurisdiction of the Tokyo District Court
- The 30-minute free allowance is presented as a promotion changeable without notice, the Private Preview API may change without notice, and payments are non-refundable except where the law requires otherwise
- The terms warn that Output is not unique and that other users may receive similar results, so treat generated code reviews, PRDs and documentation as drafts to be checked rather than finished work; the same section also prohibits automated extraction of data or Output from the Service
- SOC 2 appears on the homepage as on the way, which means it has not been obtained, and the privacy policy dated 1 October 2024 has not been refreshed alongside terms revised on 21 October 2025
Setup & Integrations
Technical difficulty
Low for the core scenario. Signing up takes an email address or a Google, Microsoft or GitHub account, the quickstart promises a first app in minutes, and the homepage claims neither code nor prompt tuning is needed. Friction appears further along: connecting the GitHub App means selecting repositories and granting rights, while Vector Stores and Data Stores require a Pro plan, with Data Stores also requiring a PostgreSQL connection. Two paths are clearly harder: self-hosting the open-source edition, which is a genuine engineering task, and the API, still in Private Preview and therefore out of reach for most users.
Deployment
Integrations
Supported languages
Behind Giselle
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
Do I need to know how to code to use Giselle?
Which AI models can I use?
Is there a free plan?
How much does the first paid plan cost?
Can several people work together in Giselle?
Is there an API?
Can I self-host Giselle?
Who owns the content I generate?
Is my data used to train AI models?
Where is my data hosted and what security guarantees are offered?
Should you pick Giselle?
Giselle is a young product from an established publisher. The domain was registered on 9 July 2024 and first archived on 19 September 2024, while ROUTE06, Inc. has been trading since 24 January 2020 and holds ISO/IEC 27001 certification. That gap explains much of what follows: the engineering and security practice look mature, the commercial packaging does not.
What the product does well is clear enough. The visual builder genuinely removes the need to write code, four model providers can be mixed inside a single workflow, and the GitHub integration, with triggers on issues and pull requests, repository commands and codebases indexed as knowledge, goes further than most no-code agent builders. The entry cost is low, with a permanent free plan and Pro at 20 USD per month, and the open-source edition offers a real way out: self-host it and the Vector Store and Data Store quotas disappear.
Two reservations matter. The first is practical: Giselle is a single-user platform today. Free and Pro each allow one user, and the Team plan that would allow ten is still Coming Soon, so every team scenario rests on a promise. The second is governance: no data processing agreement, no Article 27 EU representative, no hosting country or region disclosed, and opt-out instructions for model training that the privacy policy points to but never publishes. A privacy policy effective 1 October 2024 also sits beside terms revised on 21 October 2025.
Recommended, then, to solo builders, engineers and small AI-native teams working on GitHub who want results quickly and can live with email-only support and a Private Preview API. European organisations processing personal data on behalf of others should wait for the paperwork, or self-host.
- Choosing a selection results in a full page refresh.
- Opens in a new window.