I built an AI chat that searches your work tools and cites its sources

Dev.to / 3/23/2026

📰 NewsDeveloper Stack & InfrastructureTools & Practical Usage

共有:

Key Points

SlopWeaver unveiled an AI chat that can search across work tools like Gmail, Slack, Linear, and Google Docs, citing every source inline with platform-colored chips.
The system uses a hybrid search stack (keyword search, semantic Voyage embeddings at 1024 dimensions, and reranking) plus entity resolution to unify identities across tools, with Claude generating the responses and numbered citations.
In the demo flow, users can open a security assessment report, get a quick summary, request deep analysis, and receive a structured report with CVSS scores, remediation status from Linear, a Slack-based supply-chain timeline, stakeholder responsibilities, and personal action items with deadlines, citing 33 sources.
The stack powering the solution includes Tauri v2, NestJS, React 19, Claude (Anthropic SDK), Supabase + pgvector, BullMQ, and Voyage AI, and the project is being built in public to show the intelligence layer behind an earlier approval workflow.

Every AI assistant I've tried needs you to manually provide context from your other tools.

SlopWeaver's AI chat skips that. It's connected to your work tools (Gmail, Slack, Linear, Google Docs, and more) and searches across them when you ask a question. Every source is cited inline with platform-colored chips. Hover to preview, click to navigate to the original.

This demo shows the flow:

Open a security assessment report in the inbox
Ask the AI "what's this about?" and get a quick summary
Ask for a deep analysis: the AI triggers workspace search, knowledge lookup, and extended reasoning
Get a structured report: findings with CVSS scores, per-finding remediation status from Linear tickets, a supply chain incident timeline from Slack, stakeholder responsibilities, and your personal action items with deadlines
33 sources cited inline from Gmail, Slack, Linear, and knowledge sources

The search layer is hybrid: keyword + semantic (Voyage AI embeddings, 1024 dimensions) + reranking (Voyage rerank-2.5). Entity resolution connects "Daniel Frost" across platforms into one identity. Claude generates the response with extended reasoning and numbered citations pointing back to specific source documents.

The citation UX was the hardest part to get right. Each citation chip is color-coded to its platform (Gmail orange, Slack purple, Linear blue). Hovering shows a preview card with the original subject, sender, timestamp, and excerpt. Clicking navigates to the source in the inbox or opens it externally.

Stack: Tauri v2 (desktop), NestJS, React 19, Claude (Anthropic SDK with prompt caching), Supabase + pgvector, BullMQ, Voyage AI.

Building in public. The previous demo showed the approval queue (AI drafts, you review before anything sends). This one shows the intelligence layer behind it. Together they tell the story: AI that can see across your tools AND still waits for you to act.