2 layers
Team chat, and a layer for prompt design
One surface for organizational talk. One for chaining and library work.
Case Study
Collaborative Conversational AI assistant with automation
A California web-annotation startup needed more than a single-user ChatGPT tab. Vstorm built an open-source, self-hosted workspace: one layer for the team, one layer for prompt design and chaining, talking to API models and custom LLMs on their own iron.
The outcome
Many people, two layers, models they can host
The published product is a collaborative LLM platform: real-time multi-user work, a prompt library, memory across turns, and a path to G-suite. We do not publish a seat count or a latency number. The client is not named.
Open-source and self-hosted were the point: security, safety, transparency, and control over models trained on company data. That is not a public chatbot wrapper.
Real-time
More than one user on the same thread
Collaboration is the product, not a later add-on.
Self-hosted
API models and custom LLMs on their infrastructure
LangChain holds the pieces. Company data stays under company control.
Client voice
Thanks to Vstorm's continued work, the feedback-driven approach has led to significant improvements in project execution and efficiency. The team is responsive and collaborative, and internal stakeholders are particularly impressed with the service provider's commitment to excellence.
CEO
Web annotation tool development company
About the client
A California startup, founded in 2011, builds open-source tools so people can talk on top of the world's knowledge — annotations on the web, not a closed forum. The founder's background is climate change. The company is not named here.
They needed an LLM workspace that matched that ethic: open, self-hosted, usable with the tools the organization already had.
Vstorm's impact
Vstorm's impact, the TL;DR
- Open-source, user-friendly LLM platform for more than one person at a time
- Dual layer: organizational chat plus prompt design and chaining
- API-based models and custom LLMs on the company's own infrastructure
- LangChain for integration, data flow, and automation
- Prompt library, memory across previous messages, G-suite path
- Safety and control over models trained on company data
The challenge
A versatile LLM platform, not another single-user tab
The brief was an AI platform that is open-source and actually usable: several people in real time, several state-of-the-art models, including custom ones on their own machines. Self-hosting was how they would keep security, safety, transparency, and control over models trained on company data — and keep the answers accurate because of that, not in spite of it.
Custom-developed LLMs and LangChain were the engineering spine, not a slide about collaboration.
How we delivered
Two layers, models they can host, a library they can grow
Vstorm still adds features. The published core is the collaborative assistant with automation.
Name the workspace the tab was not
Workshops on real-time multi-user work, self-hosting, and which models — API and custom — had to sit behind one surface. Annotation culture does not fit a closed vendor chat.
- Multi-user requirement
- Self-host constraint
- Model mix
Proof of Value on the dual layer
One layer for organizational communication. One for prompt design and chaining. A prompt library. Memory so the next turn still knows the last. LangChain to keep those pieces in one flow.
- Dual-layer prototype
- Prompt library
- LangChain integration
Ship, then keep adding under feedback
Safety controls for custom LLMs. A path to G-suite so the workspace is usable on day one. Continued feature work after the first release — the CEO's published note is about that ongoing loop.
- Self-hosted path
- G-suite integration
- Feedback-driven extras
How it works
Talk as a team. Design the prompt on a second layer.
Users collaborate in real time. The organizational layer is for people. The design layer is for prompts and chains. Memory carries context from earlier messages. Diverse LLMs plug in — hosted APIs or custom weights on their infrastructure.
Team in the workspace Real-time, more than one user
↓
Organizational layer Communication for the people in the room
↓
Prompt-design layer Chaining, library, automation
↓
Models behind both API or custom, self-hosted when it must be
Undisclosed California startup × Vstorm — collaborative LLM workspace
Results
A workspace they can keep, not a chat tab they rent
Vstorm continues to add features to the Conversational AI assistant. The design layer is where new prompt ideas live. Users get transparency and a collaboration path that does not dump company text into a public model by default. We do not publish a user count. The published facts are the dual layer, self-hosting, LangChain, and a feedback loop the CEO was willing to put on the record.
Work with us
Ready to see how agentic AI transforms collaborative LLM workspaces?
Meet directly with our founders and PhD AI engineers. We will walk through real implementations from 30+ agentic projects and the practical steps to integrate them into your workflows.