A seed round in a lab built to ship.
Gargantua Labs builds the core systems AI needs — foundational technology for memory, adaptation, control, and security. We’re raising a seed round to scale engineering on DeltaWrite, expand integrations with frontier-model partners, and ship the next core systems in the pipeline. Open round; no commitments yet; conversation-first.
RoundSeed
TargetTBD
ValuationOn call
InstrumentPost-money SAFE
StatusOpen
§01 — The thesis
What you’re actually funding.
The shortest honest version of the bet. Read it twice; we’ll walk you through the longer version on the first call.
Execution is collapsing toward zero. Any team with an API key and a weekend can now build software that would have needed a funded startup a decade ago. What still compounds is product judgment in domains where the technical edge is real and the moat is in the execution path, not the headline idea.
We build the layer between “change the prompt” and “retrain the model” — the foundational technology a deployed AI system needs but today doesn’t have. The roadmap targets four core systems: memory, adaptation, control, and security. Our first product, DeltaWrite, is the adaptation layer — bounded persistent runtime adaptation that keeps the base model bit-identical on unrelated traffic. In MVP today, with deep technical work that small operator teams can ship faster than large ones. More core systems are in the pipeline.
The model is capital-efficient by design. A small AI-augmented team plus compute, audits, and partner-led GTM is the operating footprint. There are no sales quotas, no enterprise integration contractors, and no large-team coordination cost. The output is a portfolio of core systems — durable infrastructure with independent monetization paths.
§02 — Products
Where the work actually lives.
One product in MVP today. More core systems in the pipeline. The underlying R&D pipeline serves all of them.
AI · 01
DeltaWrite — adaptive inference.
Persistent, reversible behavioral overlays on frozen language models — without fine-tuning, without per-query prompt cost. Validated on six base models (0.5B–72B) across three architectural families. Production use cases include knowledge install and runtime safety patches against jailbreak attacks. MVP shipping to integration partners.
AI · 02
Knowledge editing & routing.
An active research thread on inserting, updating, and routing between modifications to a deployed model. Could surface as a DeltaWrite extension or as its own MVP; held internal until that decision is forced.
AI · 03
Founder leverage.
One operator, AI-native workflow as leverage, obsessive build cadence. Two decades of building software companies; now a small lab that ships durable infrastructure. The asset on the cap table isn’t a pipeline — it’s a founder picking technical bets at the core-systems layer and shipping them.
AI · 04
Evaluation under adversarial load.
Benchmarks and protocols that stress-test the claims a system actually makes — paraphrase robustness, threat-model coverage, capacity stress, audit hygiene. Every product ships with this discipline.
§03 — What exists today
The current state of the lab.
Written honestly. The lab was founded in 2026; the footprint reflects that. These numbers will move — this page will move with them.
Products in MVP1
DeltaWrite shipping to integration partners. Additional core systems in the pipeline.
Pipeline statusLive
The internal R&D pipeline is running. Daily human-directed sessions feed the swarm.
Team1
Nathan Peterson, founder & CEO. Two decades of startup experience across software and AI.
RunwaySelf
Self-funded to date. The seed round is the first external capital.
Revenue$0
Pre-revenue. DeltaWrite pilots are unpaid integration partnerships.
Advisory benchOpen
Seats open for AI research advisors, inference systems advisors, operator advisors, and capital advisors.
§04 — Use of funds
Where the seed capital goes.
Directional only; exact mix negotiated against the final round size. Most of the capital goes to engineering bandwidth and the next core systems in the pipeline — the line items that determine shipping cadence.
§05 — Risks, candidly
The things that could sink this.
A one-person lab in a volatile domain. The honest list, ranked by severity, with current mitigations.
Severity · High
Stage risk
One person. One MVP. No revenue yet.
The lab is pre-revenue, single-operator. DeltaWrite is technically real and shipping at MVP, but doesn’t have a closed paid contract yet. The first paying customer is the event that de-risks the model — and it hasn’t happened.
Capital efficiency buys time: the lab can run on a small burn until that event. Seed funding is sized against that reality, not against aspirational cadence. DeltaWrite has live integration partners; the technical work is documented and reproducible.
Severity · Medium
Key-person
Key-person risk.
The lab is one person. A serious interruption to the founder slows or stops shipping cadence on every system simultaneously.
The pipeline and product are documented and reproducible; the hard knowledge lives in code and prompts, not only in the founder’s head. The seed round budgets for early engineering hires that hold a second point of continuity.
Severity · Medium
Market timing
AI infrastructure cycles.
Frontier-model economics move fast. A sharp downturn in AI infra spend could compress DeltaWrite’s addressable market. A radical platform shift (e.g. away from frozen-model deployments toward continuously trained ones) could change what systems operators actually need.
Capital efficiency means the lab can survive a pause in market spend for several quarters without a forced raise. The core-systems layer is the one most likely to remain relevant across platform shifts — runtime control, safety, and adaptation are needed regardless of the training paradigm.
Severity · Low
Model portability
DeltaWrite host-family regression.
DeltaWrite has been validated on six base models across three architectural families. A new frontier-model architecture (e.g. a radical departure from transformer attention patterns) could require materially new calibration and break the current portability story.
The construction is parametrized over architecture; per-family calibration is treated as part of the deployment work. Active monitoring of frontier releases; new families brought into the validation matrix as they ship.
§06 — Start a conversation
One call. No memo required.
We prefer a 30–45 minute conversation to a one-shot deck review. Bring your sharpest questions; we’ll walk through the thesis, the product, the pipeline, and the risks. If there’s fit on both sides, we move to a term sheet inside two weeks.
- Contact
- Owner
- Nathan Peterson
- Deck
- On request
- Data room
- After first call
- SLA
- One business day