Startup profile · funding coverage
Sqwish Labs funding, valuation and investors
Cambridge AI startup making small, tunable decision models for agent actions, classification and risk gates through its OneHot platform.
Verified company research · funding, investors and latest developments
Keep track of Sqwish Labs
Save this profile to your VCT watchlist for a quick return.
Funding, valuation & investors
Answer-first snapshotLatest funding
Pre-seed
$3M total round incl. prior converted holdings · November 18, 2026
Latest known valuation
Not publicly disclosed
Total disclosed equity funding
$3M
Excludes debt, grants, acquisitions, secondaries, and IPO proceeds.
Current status
private
pre seed · Cambridge, United Kingdom
Investors in latest funding
Lead: Cambridge Enterprise Ventures
Other: Parkwalk , Steve Young , Karthik Tadinada , Vento
Overview
Founded in 2024 by University of Cambridge alumni Dr Ushnish Sengupta (CEO) and Federica Freddi (CTO), Sqwish Labs builds the kettei family of fast, specialized decision models and the OneHot training and hosted-inference platform. Its $3M pre-seed was led by Cambridge Enterprise Ventures, with Parkwalk, individual angels and research support from Innovate UK; some earlier investors converted prior investments into the round.
The investor announcement describes the $3M as total funding; do not treat earlier notes as additional cash on top. Valuation, audited ARR, margins, commercial retention and independently verified comparative savings were not disclosed. Company says its models ranked highly on JevBench and ImageJevBench as of Oct 29, 2026; benchmark rankings are task- and date-specific and do not demonstrate safe production use in every domain.
Why Sqwish Labs is interesting
Cambridge Enterprise Ventures led a $3M pre-seed round reported November 17–18, 2026, for AI systems that choose among possible actions instead of generating prose. Company publishes unusually specific token, training and keep-warm prices; the key unanswered question is whether benchmark wins generalize to real customer workflows, data drift and safety thresholds.
Product & use cases
Illustrative customer-support-operations workflow drawn from docs: (1) define a bounded choice such as 'escalate / don't escalate' and specify required evidence; (2) route ticket text or supported images to kettei through OneHot API; (3) receive option probabilities rather than a generated paragraph; (4) apply a confidence threshold, with ambiguous examples sent to a larger model or human reviewer; (5) label correct and incorrect decisions and fine-tune a supported model on customer examples; (6) compare candidate against holdout cases, deploy with monitoring, and roll back when quality falls. Models fat/chub/core have listed hosted prices of $0.06/$0.04/$0.02 per million input tokens; no output-token charge, prepaid credit, before tax. Fine-tuning: chub $0.80 and core $0.40 per million training tokens, minimum $2.50 per successful job. Optional keep-warm $0.10/hour, billed for reserved capacity. No fixed subscription fee. Model-specific capabilities and preview status apply; two small models have open weights. Vendor-defined demo accuracy and efficiency are not third-party customer trials.
- Escalating or routing support tickets and agent tool calls based on probabilities
- Contract clause classification and expert-defined legal questions, with human review for material decisions
- Filtering prompt-injection attempts and other bounded safety decisions, subject to independent validation
- Visual classification with kettei fat/chub on supported image inputs
Key facts
- $3M Cambridge Enterprise Ventures-led pre-seed disclosed Nov 17–18, 2026; prior convertible investments included, exact closing date undisclosed.
- Founder and co-founder: Ushnish Sengupta and Federica Freddi, Cambridge-based, established 2024.
- Prices verified against vendor docs updated Nov 15, 2026: core $0.02, chub $0.04, fat $0.06 per 1M input tokens, with additional fine-tuning/keep-warm charges.
- JevBench/ImageJevBench performance refers to specific October benchmark snapshots; vendor examples are not independent customer ROI.
Funding history (newest first)
Pre-seed
2026-11-18 $3M total round incl. prior converted holdings- Cambridge Enterprise Ventures (lead)
- Parkwalk (participant)
- Steve Young (participant)
- Karthik Tadinada (participant)
- Vento (participant)
Investors in our directory
Funds linked from Sqwish Labs's profile — open a fund page for stage focus and related deal articles.
Competitive landscape
Edge: Single-purpose, quickly calibrated probability decisions with fine-tuning and open-weight options at small-model sizes, enabling hybrid pipelines that reserve expensive general-purpose models for hard cases. This trades away rich free-form language and requires well-designed labels and confidence calibration.
Evaluate measured decision precision/recall, calibration and false positives on real private examples; compare training data rights, hosting/residency, portability of fine-tunes, fallback costs and threshold effects. Top benchmark placements are not proof of better real-world legal decisions or security outcomes.
-
TypeSafe Jev direct
Purpose-built decision-model alternative; benchmark and API compatibility must be checked on a customer's own data.
-
OpenAI decision models direct
Vendor alternative to benchmark for performance, cost, governance and supported modalities.
-
Microsoft decision models direct
Another announced decision-model option; production feature coverage and commercial terms require direct comparison.
-
General-purpose LLM with JSON classification alternative
More flexible free-form reasoning, potentially higher latency/cost for simple bounded decisions.
-
Custom classifier or rules engine alternative
May be easier to audit for stable taxonomies; needs explicit training and upkeep if business rules drift.
Notable stories
- Lead investor Cambridge Enterprise announcement: enterprise.cam.ac.uk (external source)
- Company product and use cases: sqwish.ai (external source)
- Vendor pricing updated Nov 15, 2026: docs.sqwish.ai (external source)
- Model specifications: sqwish.ai (external source)
- Company experiment: sqwish.ai (external source)
Market / IPO context
Editorial / static context — not a live quote.
Last updated:
Editorial note: AI tools assisted with research, structure, or drafting. Venture Capital Tracker retains human editorial responsibility for factual accuracy, relevance, and source quality before publication.