Kin Pilot

Open Source Coming Soon Rust Apache-2.0

Kin-Native Agent Shell

Today's AI coding assistants are deeply trained on file-system mental models — they think in paths, diffs, and directory trees. Kin Pilot retrains that reflex. It makes the code agent natively understand and prefer Kin's semantic graph, so it reasons about functions, types, and relationships instead of lines and files.

Coming Soon Back to Labs

The Problem

Every AI coding assistant today — Copilot, Cursor, Claude Code, Codex — thinks in files. They cat files, grep for patterns, and diff line numbers. That's because they were trained on file-system interactions. Kin Pilot breaks that assumption.

File-System Thinking

Traditional assistants read entire files, guess which ones matter, waste tokens on boilerplate, and lose track of how code connects across boundaries.

Semantic Thinking

Kin Pilot queries the semantic graph directly — it knows what a function does, what calls it, and what it depends on. No file scanning required.

Native Kin Preference

The agent doesn't just support Kin — it prefers it. When a semantic graph is available, the agent reaches for entities and relationships first, falling back to files only when necessary.

Graceful Fallback

Works alongside existing Git workflows. In repos without Kin, it behaves like a standard agent shell. In Kin-initialized repos, it unlocks semantic superpowers.

How It Works

Developer Asks

Developer asks a question or requests a code change through the agent.

Queries Semantic Graph

Kin Pilot queries Kin's semantic graph to find relevant entities and relationships.

Token-Budgeted Context

Gets precise, token-budgeted context — only the code and relationships that matter.

Informed Response

Generates an informed, accurate response grounded in real codebase structure.