New Projects Radar

New Projects Radar: From Daily Discovery To Durable Memory

This page defines how we continuously scan new projects and turn raw findings into reusable memory for teams.

Goal

1. Detect relevant new projects quickly (tools, frameworks, papers, launches).
2. Store the signal as tenant-scoped memory so agents can cite it later.
3. Keep the loop cheap and automated: cron + optional manual refresh.

What Gets Ingested

• Product and launch signals (Hacker News, vendor blogs, release feeds).
• Open-source movement (project announcements, major release notes).
• Research-to-product signals (selected arXiv queries for agentic tooling and software workflows).

Memory Shape

Daily digest memories are stored as:

• category: research
• tags: research, new-projects, daily, <YYYY-MM-DD>
• title: New Projects Daily Digest <YYYY-MM-DD>
• context.kind: new-projects-digest
• context.items: normalized source entries (title, URL, publish date, source)

This keeps retrieval simple for both the web app and the MCP layer.

How To Use It In Agents

1. Query recent new-projects digests for the current tenant.
2. Extract candidate tools/projects by relevance to the active project memory.
3. Promote high-signal findings into explicit project memories:
   • migration ideas
   • architecture options
   • risks and adoption notes

Iteration Rules

• Prefer reliable machine-readable feeds over brittle scraping.
• Keep the feed list small and auditable; rotate sources by measured signal quality.
• Keep digests evidence-first with direct links.
• Add source-specific confidence later (e.g., official docs > social posts).

Next Up

1. Add dedupe across days (same URL tracked as an update, not a new memory).
2. Add relevance scoring against current project stack tags.
3. Add "promote to project memory" action directly in the digest UI.