Browther – AI code search filter CLI for coding agents

Deliver more precise code retrieval for coding agents: recall then filter to save context

Problem

When feeding code snippets to coding agents, directly stuffing semantic search results wastes context and adds noise. A V2EX user built the sj CLI, using Jev to filter Semble results, proving this is a direction to optimize agent workflows.

Solution

A CLI tool that combines local semantic retrieval (e.g., Semble) with lightweight LLM filtering to return curated code snippets, file paths, and line numbers to agents. Filtered-out snippets retain positions for follow-up reading. Integrates as an agent skill.

Users

Developers using coding agents, AI toolchain builders, teams needing efficient code search.

MVP

CLI command to query current project; integrate Semble or similar local search; call lightweight LLM (or rules) to filter results; output compact JSON for agent; configurable filter threshold.

Revenue

Open-source core, enterprise version with team-shared index, audit logs, and priority support.

Why now

AI coding agents are widespread, context windows remain a bottleneck, and precise retrieval significantly boosts agent efficiency. Indie developers can quickly build on existing open-source components.