interject

interject

Ambient discovery and research engine for Claude Code.

What this does

interject continuously scans arXiv, Hacker News, GitHub, Anthropic docs, and Exa for new capabilities, tools, and research relevant to your projects. It builds a learned interest profile from your promote/dismiss signals and gets better at surfacing relevant discoveries over time.

When something interesting turns up, it creates a briefing with context about why it matters for your specific projects: not just "here's a new paper" but "here's a new paper and here's how it relates to the embedding infrastructure you're building in intersearch."

High-relevance discoveries get promoted to brainstorm docs and beads. Medium relevance gets briefings. Everything else goes in the digest. The confidence tiering means you're not drowning in noise.

Installation

First, add the interagency marketplace (one-time setup):

/plugin marketplace add mistakeknot/interagency-marketplace

Then install the plugin:

/plugin install interject

Requires intersearch as a dependency (shared embedding infrastructure).

Usage

Scan for new discoveries:

/interject:scan

Check your inbox:

/interject:inbox

View and manage your interest profile:

/interject:profile

Architecture

src/interject/       Python library with pluggable source adapters
skills/              scan, discover, inbox, profile, status
server/              MCP server (Python/FastMCP, launched via uv run)

SQLite database tracks discoveries, promotions, feedback signals, and query history. The recommendation engine uses scored ranking with feedback-driven weight adjustments.