About Kin
AI writes code. Kin proves the change. Kin is the system of record for AI-written software: Git for the agent era. It runs beside Git and CI today as a report-only pre-merge evidence layer.
The wedge
The expensive failure is rarely in the file that changed. Files compile locally; systems break relationally.
Most tools in the category layer AI on top of the file-and-diff substrate: Copilot, Graphite, Sourcegraph, Cursor, Greptile. Kin replaces the substrate. Code lives as a living semantic graph, and review, agents, and collaboration all read from that graph instead of re-deriving structure from text.
Why now
AI writes a fast-growing share of the code that ships. But the repo, diff, and review substrate underneath it was built for one thing: humans editing files, one line at a time. That model is straining.
When an agent refactors across a dozen files, a unified diff hides the one change that quietly breaks an assumption three modules away. When many agents work at once, a file-locking, branch-merging workflow has no idea what they each intended. Establishing impact and review evidence has become the bottleneck. Kin records that evidence so machine-written code can be reviewed with context.
The dual thesis
KinLab has two jobs at once. It is the entrypoint to the open Kin tooling, the easiest way to adopt the semantic substrate and wire it into your editor and your agents. And it is the hosted control plane, where teams govern agent access, record review decisions, and review change by impact across an org-wide graph. Automatic merge blocking is not the default today.
Open tooling earns adoption; the hosted layer is the business. The two reinforce each other: the more the open substrate spreads, the more valuable the shared, governed layer on top becomes.
The stack
One system, a few clear surfaces, from graph-owned truth at the core to hosted collaboration on top.
Semantic system of record
The graph-native repo format, CLI, daemon, and review/provenance semantics. Graph-owned truth is the source of authority.
Transparent filesystem projection
An optional virtual filesystem that serves graph-backed files to compatible tools on glibc Linux and macOS, subject to host constraints.
Editor and agent surfaces
A lightweight editor extension and the MCP server bundled in Kin give humans and supported AI agents graph-native search, context, and edits.
Hosted collaboration
The proprietary hosted KinLab control plane for teams, org memory, governance, and report-only review, currently in early access.
Open at the core
Open-source Kin substrate, hosted KinLab control plane. Adopt the tooling for free, scale on the platform. The Kin substrate is open source under Apache-2.0: semantic version control, the platform-limited VFS, the bundled MCP server, the editor, and the engine. The local core runs without a hosted account; VFS is optional and host-dependent. Features that ship open never move to a paid tier; the moat is the hosted control plane and your org’s graph. See what’s open.
We build Kin with Kin
A small team directs fleets of AI agents against our own semantic graph, using the same governance, review, and context surfaces we’re shipping. The development process is the product thesis: if a few people can safely steer many agents, it’s because the substrate makes that possible.
Where this is today
Open-source Kin v0.3.4 is the current public developer-alpha release. The hosted KinLab workspace is separate, proprietary, and remains early access by request. We keep our claims to what the published evidence supports. You can read the public proof and browse the components in the open Kin ecosystem.