CompletedLangGraphTypeScript
Linkwizard AI
LangGraph AI assistant that lets teams query and manage internal bookmarks in Slack using natural language.
Timeline
Ongoing
Role
Developer
Team
LabsGPT
Status
CompletedTechnology Stack
LangGraph
TypeScript
Key Challenges
- Routing natural language to the right tools
- Giving the LLM strict, reliable schemas
- Improving retrieval on loose queries
Key Learnings
- LangGraph orchestration
- API client design for LLMs
- Query rewriting
Linkwizard AI
A LangGraph assistant that lives inside Slack, so teams can query and manage their internal bookmarks in plain language instead of digging through a UI.
Overview
The bot connects to a self-hosted Linkwarden instance and lets the LLM run real database operations through a tightly-scoped API client.
Technology stack
Technology Stack
LangGraph
TypeScript
Project details
Timeline
Ongoing
Role
Developer
Team
LabsGPT
Status
CompletedA strict API client
Reliability comes from giving the model strict schemas and system instructions rather than open-ended tools:
// The LLM only sees well-typed operations
const tools = [
{ name: 'searchBookmarks', input: { query: 'string' } },
{ name: 'createBookmark', input: { url: 'string', tags: 'string[]' } },
];Better queries before better tools
Natural language is rewritten before it reaches the tools. Query rewriting and multi-query searching expand a single request into parallel searches, improving retrieval accuracy.
Key challenges
Key Challenges
- Tool routing accuracy
- Keeping schemas strict
- Handling imprecise queries
What I learned
Key Learnings
- LangGraph state machines
- Designing LLM-friendly APIs
- Query rewriting techniques