Giselle

Build and run AI workflows. Open source.

Productivity

Open Source

Artificial Intelligence

Oh, we’re actually working on a couple of AI startups in tourism and e-commerce. We’ll check out your product, thanks.

Great work guys! It looks amazing! I was wondering what is the basis of your intelligent model selection? What algorithm do you use, if you don’t mind me asking? Does the model router optimize for things such as 1) execution cost, 2) input tokens cache hits, 3) speed? Thanks in advance.

This resonates. Visual workflows only become useful once debuggability and execution transparency are first-class, not afterthoughts. Open source plus real-time visibility is a strong combination, especially for long-running, multi-step flows where trust is built by being able to see what ran, why it ran, and where it failed. Curious to see how this evolves as teams push it beyond experimentation into shared production workflows.

You’re basically packaging “LLM orchestration as a legible object” turning invisible glue code into something humans can reason about.
Knife-edge question: what’s your plan for reproducibility (snapshots of models/inputs/connectors) so workflows don’t drift into “works on my Tuesday”?
I love this because I hate myself. (Kidding. Mostly.)

This is really well-executed. The visual node editor for multi-model workflows is exactly what the agent ecosystem needs right now.

Quick question: How do you handle state persistence across long-running tasks? We're launching Nex Sovereign tomorrow (cognitive OS with persistent memory + governance layer) and thinking about how workflow builders like this could interface with stateful agents.

Upvoted! 🚀

Looks very simple to use. I'm just wondering if you can build a workflow by writing a prompt, and will actually come up with the required nodes and suggest integrations?

This is mind-blowing! 🤯 The visual builder for chain-of-thought agents is exactly what the dev ecosystem needed.

I'm really curious about the 'GitHub-native' RAG part. How do you handle context limits when indexing massive repositories? Do you have intelligent chunking for code specifically?

Congrats on the launch, team!

Looks cool! Thx for you guys build a easy-used workflow, which is really clear with input and output

Hits a nerve tbh. I’ve bounced off a bunch of workflow tools—too much setup, no clue when jobs hang. Open source + real-time view + mix providers sounds right. Gonna try it on a long-running scrape/summarize flow. Curious about retries/state.

Congrats on the launch! Building something you actually wanted to use really shows in how you’ve framed the problem. The focus on visibility and debuggability especially hits that’s where most workflow tools fall apart in real use.

I appreciate the open-source approach here. It makes it feel less like a black box and more like a tool I can actually trust and grow with.

Congrats on the launch, the focus on clarity and flexibility in a workflow really stands out.

Congrats on the launch! "Zero infra setup" is exactly what I've been looking for—most AI workflow tools are way too heavy or complex to just get started. The visual node editor looks super intuitive. Since it's open source, are you planning to add community-contributed nodes/integrations soon?

Hey Tadashi, that line about workflows being too opaque to debug really hits. Was there a specific one that broke and you just couldn’t figure out why?

I'm Taka, CEO of the team behind Giselle. Today we're launching Giselle — a visual AI app builder designed for product teams.

Why we built this:

We started building Giselle over a year ago, back when GPT-3 was the standard and tools like CrewAI and n8n were just emerging. Our original goal was to bring LLM-powered automation to consulting and finance — domains drowning in research and documentation work.

But here's what we learned: getting "professional-grade" output quality was hard. Really hard. So we did what any stubborn team would do — we dogfooded relentlessly. We became our own zero-customer, using Giselle daily to build our own products.

That journey shaped what Giselle is today: an AI app builder optimized for product ops and GitHub-native workflows.

What makes Giselle different:

• GitHub as your vector store — Turn your repos, issues, PRs, and code into RAG-ready context with one click. No pipeline setup.

• Event-driven workflows — Trigger Giselle apps from GitHub events (new issue, PR comment, etc.). Build your own CodeRabbit-style review agent — no code required.

• Team-first, cloud-native — Apps you build are instantly shareable. Call them from a chat UI ("Stage") or directly from GitHub with custom slash commands (you define the command name).

What you can build:

• ✅ Automated PR review agents

• ✅ PRD drafters that pull context from your codebase

• ✅ Spec/docs updaters triggered by merged PRs

• ✅ Parallel workflows like Cursor or Claude Code — but for your whole team

Why this matters:

Tools like Cursor and Claude Code have supercharged individual developers. But teams still struggle to share that leverage. Giselle bridges that gap — not everyone needs to be a builder; one person's app becomes the whole team's productivity boost.

What's next:

Right now we're focused on product ops, but the path forward is clear. As we expand support for diverse document types and data sources, we expect Giselle to handle the consulting and professional services use cases we originally envisioned — research synthesis, client deliverables, knowledge management at scale.

In the near term, we're doubling down on two fronts:

• Visual builder improvements — Making it even easier to prototype AI apps without code

• Developer-facing features — Instant API access for any app you build, MCP (Model Context Protocol) support, app virtualization for complex compositions, and smoother paths to scale with LangChain when you're ready

We'd love your feedback. What workflows would you build first?

Hi Product Hunt 👋

I’m one of the creators of Giselle. Thanks for checking us out.

When we looked at existing AI workflow tools, they mostly fell into two camps.

Some are extremely easy to start with. You can build something quickly and see results right away — but once you try to use them for real work, they start to feel fragile. Long-running jobs and failures make you nervous, and it’s hard to understand what’s happening while they run.

Others are built for durable execution, with strong telemetry and observability. They’re powerful and reliable, but the first step is heavy, often requiring engineering effort before you can even try an idea.

We couldn’t find something that was easy to start and still safe to rely on.

That gap is exactly why we built Giselle.

Giselle lets you design AI-powered apps and workflows intuitively, while treating them as real jobs — with clear progress, structure, and visibility, even when they run for hours across many steps.

If you’ve ever felt that your AI workflow works “until you actually need to depend on it,” we’d love to hear your thoughts and feedback.