A code repository
built for AI.
A graph-native code repository for AI agents
and the people who build with them.
AI is backing teams into a corner.
AI tools generate more code. Teams still have to understand it, check it and keep it running.
Review
+442%
median time
in review
Faros, April 2026
Unreviewed
+31%
pull requests merged
with no review
Faros, April 2026
Production
+125%
monthly production
incidents
Faros, September 2026
The bill
39%
of enterprises report
higher-than-expected
AI coding costs
Mavvrik, July 2026
Faros: 22,000 developers across 4,000 teams. April compares low and high adoption. September studies deeper use among high-adoption teams. Findings are correlational. Mavvrik and Benchmarkit: 396 enterprises, surveyed April–May 2026. Cost overruns are self-reported.
2
We keep paying to rediscover
how code connects.
Without a shared record, the next person or agent may have to trace the same connections again.
Cart total
?
Checkout amount
?
Receipt line
?
One small change
Discount rule
−Max discount 100%
+Max discount 50%
76%
of tokens went to reading in one coding-agent study.
Claude Sonnet 4.5 on SWE-bench Verified.
Illustration based on Kin's saved discount example, September 16, 2026. Dashed paths mark connections to check. SWE-Pruner, 2026: read operations were 76.1% of tokens in this study, not a measure of avoidable reading or human review time.
3
Kin versions code and its connections
so people and AI can reuse the answers.
The discount rule's recorded connections answer the same question at the next handoff.
What uses the discount rule?
Answered from
Kin's record
One question. No re-reading.
+442% time in review
Faster review
+31% unreviewed changes
More code checked
+125% monthly incidents
Earlier risk checks
39% report AI cost overruns
Less repeat reading
From Kin's saved example of September 16, 2026, on Kin 0.7.19. Kin can miss connections. The bottom row is what Kin is designed to improve; the benchmark on the next slide measured tokens.
4
Kin answered the same questions
with 91% fewer tokens.
A sealed test on held-out questions in React and Flask. Kin supplies the map, so AI can focus on the work.
91%
fewer tokens
Total prompt + completion tokens
Searching files
849,673
With Kin
79,553
79%
Correct on the first response.
No follow-up query.
19 of 24 runs
0
LLM tokens for Kin's lookup.
The graph supplies the answer.
The AI response still used tokens.
8 sealed questions × 3 seeds = 24 paired runs; all 24 Kin final answers scored correct. React + Flask; same Qwen3.8-27B model. Read-only, unreleased answer-first build, tuned earlier and frozen. Kin's answer entered the first prompt. Uncached tokens; no speed or customer-bill result.
5
A team of 200 developers could gain
~$750K in annual value.
Illustrative annual scenario for a team using AI, before KinLab fees.
Review
$520K
Engineer time
returned
Unreviewed
$45K
Less rework
after merge
Production
$45K
Less incident
response work
The bill
$137K
Lower AI
usage spend
As AI use grows, so can the value.
Beyond the estimate: fewer mistakes, stronger customer trust and retention,
and more time to deliver value.
Annual assumptions for 200 developers. Review: 6 h/week × 46 weeks × 10% saved. Rework: 10 fixes/month × 4 h. Incidents: 1/month × 40 h. Labor: $94.47/h. AI: $300/dev/month in variable usage × 19% assumed savings. Hours do not overlap. Sources: Bosu and Carver, BLS, Anthropic, LinearB. All reductions are hypothetical.
6
Kin is free.
KinLab is the paid team platform.
The first core pricing proposal is a subscription per developer seat.
Kin
Free
The open foundation
Run it yourself.
For individual developers and teams.
Your code, context and history.
KinLab
Proposed core plan
$40
per developer seat
per month
The managed team platform
Shared code context
Coordinated reviews and releases
Team and agent access controls
Potential expansion: hosted agents and automation usage, optional add-ons and enterprise packages.
Pricing hypothesis. KinLab is in development. $40 per developer seat/month is the proposed core subscription; packaging, included usage and willingness to pay require validation. Expansion is planned and separately priced. Related models: Cursor, Devin and CodeRabbit.
7
Kin can grow from developer use
or a leadership mandate.
First customers: teams of 20 to 200 developers building with AI agents on large codebases.
DEVELOPERS BUILD MOMENTUM
LEADERS DRIVE ROLLOUT
3
The organization sees value
A shared answer to cost,
review and coordination.
2
Developers share it
Teammates try it
in their own work.
1
One developer starts
Less repeated investigation.
More useful AI context.
1
Leadership feels the pain
AI bills, delivery friction
or risk demand action.
2
CTO / CISO sets a standard
A leadership-backed pilot
becomes a rollout.
3
Teams adopt together
A shared foundation
across the organization.
Kin
One shared foundation
Start with one team.
Build repeat use. Expand with KinLab.
Proposed adoption strategy. Kin is free and in public beta; KinLab is in development. Developer-led sharing and leadership-led rollout are paths to test, not demonstrated customer traction.
8
FOUNDER
Troy Fortin Jr
Founder and creator of Kin. Full-time on Kin when the round closes.
I built Kin after spending my own money on AI that kept losing context and sending me back to fix its work.
Hands-on experience
AI engineering at The Home Depot
Built from that experience, self-funded, by directing coding agents
Kin, in public beta
123
releases since June
14
languages supported
17,600+
automated tests
9
Kin puts code connections
inside the repository's own history.
The map becomes part of the foundation people and agents build on.
SELECTED EXISTING APPROACHES
Extend the existing
repository model.
Each adds a layer on top of it.
Code hosts
Context engines
Agent records
GitHub
GitLab
Sourcegraph
Augment
Entire
Agent Trace
Supported by Cognition
We believe more layers are a costly, short-term answer.
Kin
Code and recorded connections,
versioned together.
Tracked content
Recorded connections
One repository history
Commit, branch and merge them together, entity by entity.
Patent pending
A versioned, tamper-evident graph of code
and its connections.
Stays current
Edits update only the connections they touch.
Kin flags anything not yet rechecked.
Works beside Git
Kin reads Git history without changing it,
so teams adopt it one repository at a time.
Architectural thesis. Products overlap in graph, context and history features; logos link to documentation. Agent Trace is an open spec. Kin architecture.
10
Our first year after funding moves
from hiring to paying teams.
Milestones and targets for the first four quarters after the seed round closes.
Hire the first team
Reliability and developer-experience engineers.
Work with design partners
Three teams, mid-Q1 through mid-Q2.
Win the first paid teams
Paid pilots from Q2, shaped by their feedback.
Expand the team
A hosted-deployment engineer in Q4.
What we raise on next: repeat team use, pilots that become production teams, and a repeatable way to win them.
Planned sequence from our operating model. Kin is self-funded to date; fundraising for the next round starts around month 15.
11
A $3.75M seed takes Kin
from public beta to paying teams.
Spending follows what each milestone proves.
SEED ROUND
$3.75M
24 months of runway
Valuation and instrument in conversation.
Product and engineering
70%
The first engineers, reliability, security reviews and hosted KinLab.
Go-to-market and community
9%
Design partners, developer adoption and paid pilots.
Operations
3%
Legal, accounting and insurance.
Reserve
18%
A three-month operating reserve, plus room to scale what works.
Superintelligence needs a better foundation.
Rounded from our operating model, available on request during diligence.
12