Kin
v0.2.18AlphaApache-2.0Semantic system of record for software work
Git stores text history. Kin understands code.
Kin replaces the file-first, diff-first repository model with a graph of semantic entities and relationships, then serves graph-backed context to developers and AI agents. It is not a coding assistant and not a Git wrapper. It is the local semantic substrate that the rest of the Kin ecosystem builds on.
Deterministic retrieval
26 / 26
Kin runs were bit-identical; the lexical baseline was bit-identical on 3 of 26 tasks.
Symbol F1
.318 vs .289
Kin leads the lexical baseline in the pre-registered n=26 evaluation.
Cost parity
1.00x tokens
Kin used 0.95x tool calls; file F1 remained a statistical tie, not a claimed win.
Scope: one pre-registered 26-task Go retrieval and localization suite on one model and runtime. This is not an end-to-end patch-success result.Read the public proof summary and governing artifacts.
What Ships Today
The public Kin story is no longer about custom forks. It is a local semantic substrate with agent, editor, and hosted layers built around it.
Semantic system of record
Kin makes the entity graph authoritative and treats files, diffs, and review flows as projections on top.
60+ tool MCP surface
Any MCP-compatible agent can query semantic search, trace, context, review, and impact directly through Kin.
Brownfield migration
Kin works with or without .git. Import existing repos, keep Git remotes, and adopt the semantic layer incrementally.
Evidence-aware review
Per-hit source evidence, provenance, impact, and review decisions expose what Kin knows and where evidence is incomplete.
No forks required
Editors can sit on kin-vfs. VS Code gets Kin Editor. Agents connect over MCP. The graph stays central while surfaces stay replaceable.
Measured, not hand-waved
In a pre-registered n=26 evaluation, Kin led symbol, span, and line F1, tied on file F1, and ran bit-identically on 26 of 26 tasks.
Quick Start
# Install Kin and run the setup wizard
$ curl -fsSL https://get.kinlab.dev/install | sh
$ kin setup
# Initialize a repo; existing Git history imports automatically
$ cd my-project
$ kin init .
# Explore semantically
$ kin status
$ kin locate "webhook retries twice"
$ kin trace AuthService
$ kin review shadow main..HEAD
How Kin Fits
Kin is the center of gravity in the active stack: below it sits KinDB, around it sit projection and UX surfaces, and above it sits the hosted collaboration layer.
Below Kin
KinDB stores graph topology, semantic metadata, signatures, fingerprints, and search indexes.
Inside Kin
One Rust binary bundles the CLI, daemon, semantic operations, parser pipeline, migration workflows, and MCP server.
Around Kin
kin-vfs projects graph-backed files to normal tools, Kin Editor gives VS Code native affordances, and KinLab builds the shared org surface above it.
GitHub Release Sync
This page reads GitHub metadata during site deploys so the version badge follows the public release instead of a hand-edited value.
What the tag represents
Current public Kin releases include the CLI, daemon, MCP server, parser pipeline, Git coexistence, and release artifacts for macOS, Linux, and Windows.
Why this is wired at build time
Production builds verify the current GitHub tag and release before publishing. A successful deploy cannot silently replace that authority with stale fallback data.