nunu.ai
Building the first multimodal agents to play and test games.
Website: https://nunu.ai/
Cover Block
PUBLIC
| Field | Value |
|---|---|
| Name | nunu.ai |
| Tagline | Building the first multimodal agents to play and test games |
| Headquarters | San Francisco, United States |
| Founded | 2022 |
| Stage | Seed |
| Business Model | SaaS |
| Industry | Media / Entertainment (Games) |
| Technology Type | AI / Machine Learning (multimodal agents) |
| Geography | North America |
| Growth Profile | Venture Scale |
| Founding Team | Co-Founders (3) |
| Funding Label | Seed |
| Total Disclosed | ~$8,000,000 |
Links
PUBLIC
- Website: https://nunu.ai/
- LinkedIn: https://www.linkedin.com/company/nunu-ai
- Y Combinator profile: https://www.ycombinator.com/companies/nunu-ai
- Crunchbase: https://www.crunchbase.com/organization/nunu-ai
- Trust Center: https://trust.nunu.ai/
Executive Summary
PUBLIC
nunu.ai builds multimodal AI agents that play and stress-test video games the way a human QA tester would. The approach sits at the intersection of two well-funded categories: generative AI tooling and game development infrastructure.
The company was founded in 2022 by Kyrill Hux, Nicolas Muntwyler, and Jan Schnyder. It went through Y Combinator's W23 batch and is headquartered in San Francisco with a small team [Y Combinator].
Its core product, accessed through a web portal called nexus, lets studios spin up AI agents that complete tutorials, hunt for bugs, and file reports back into Slack or Jira without game-specific scripting [nunu.ai, retrieved 2024]. Public demonstrations include an agent that completed Hogwarts Legacy's roughly 30-minute opening tutorial unaided. A separate run at Pokémon Emerald was live-streamed [nunu.ai, retrieved 2024].
Capitalization stands at roughly $8 million across a 2024 pre-seed and a $6 million seed in March 2025. The seed was co-led by a16z speedrun and TIRTA Ventures [Crunchbase, March 2025] [StartupTicker].
The business model is SaaS, with tiered plans aimed at studios from indie up to AAA. A separate enterprise track includes a dedicated CSM and 24/7 support [nunu.ai, retrieved 2024].
Over the next 12 to 18 months, key signals will be paying-customer disclosures from named studios. Watch the agent's generalization across engine types and genres. Early experiments in physical robotics (the "Rob the Robot" project) may translate into a second commercial wedge or remain a research showcase [nunu.ai, retrieved 2024].
Data Accuracy: GREEN -- Confirmed by Crunchbase, Y Combinator, StartupTicker, and the company's own site.
Taxonomy Snapshot
| Axis | Value |
|---|---|
| Stage | Seed |
| Business Model | SaaS (tiered + enterprise) |
| Industry / Vertical | Game development tooling / QA automation |
| Technology Type | Multimodal AI agents |
| Geography | San Francisco, USA |
| Growth Profile | Venture Scale |
| Founding Team | 3 co-founders |
| Funding | ~$8M total disclosed (Seed, March 2025) |
Company Overview
PUBLIC
nunu.ai was incorporated in 2022 by Kyrill Hux, Nicolas Muntwyler, and Jan Schnyder. It joined Y Combinator's Winter 2023 batch [Y Combinator] [LinkedIn].
The company has described its mission as integrating AI into the game development cycle. This begins with quality assurance and extends toward live player simulation [nunu.ai, retrieved 2024].
Public positioning has narrowed from a broad "gameplay agents" framing to QA automation for studios. The broader research narrative (agents that can play any game, in any environment) remains forward-looking.
The early commercial thesis was validated in two stages. A 2024 pre-seed of roughly $2 million provided initial runway [Mobidictum, retrieved 2026].
In March 2025, a $6 million seed round closed. It was co-led by a16z speedrun and TIRTA Ventures, with participation from Y Combinator and others. Total disclosed capital reached approximately $8 million [Crunchbase, March 2025] [StartupTicker].
The company is registered as Swiss-origin but relocated its center to San Francisco. Press described it as a "Swiss startup," while databases and YC list San Francisco headquarters [StartupTicker] [Y Combinator].
Key milestones include the December 2024 Hogwarts Legacy tutorial demonstration. A follow-on was the live-streamed Pokémon Emerald run. The "Rob the Robot" project tested virtual-to-real generalization [nunu.ai, retrieved 2024] [FOV Ventures, retrieved 2026].
Headcount remains small: five on YC's profile, 1 to 10 on FOV Ventures [Y Combinator] [FOV Ventures, retrieved 2026].
Data Accuracy: GREEN -- Confirmed by Crunchbase, Y Combinator, StartupTicker, and the company website.
Product and Technology
MIXED
The product is a SaaS platform branded nexus, sold to game studios. Non-technical team members author and run test plans that an AI agent executes against a build [nunu.ai, retrieved 2024].
Every test produces a gameplay recording with the agent's reasoning trace. It includes exact keyboard, mouse, or controller inputs, error logs, and a structured report. Bugs push to Slack or Jira automatically [nunu.ai, retrieved 2024].
The headline reference is StormForge. There, test creation, agent execution, and ticket routing form a full loop [nunu.ai, retrieved 2024].
Technically, agents are multimodal. They perceive games through pixels and audio, not engine internals. They emit human-equivalent inputs [Y Combinator].
This choice lowers onboarding friction. No per-engine SDK integration is needed for proprietary or older engines.
The Hogwarts Legacy tutorial run is the most cited proof: 30 minutes without prior training or scripting [nunu.ai, retrieved 2024]. Pokémon Emerald was a stress test of long-horizon planning [nunu.ai, retrieved 2024].
The "Rob the Robot" project applied the architecture to a physical robot. It tested navigation of real space, object location, verbal commands, and home automation interaction [FOV Ventures, retrieved 2026].
The company and TIRTA Ventures describe this as evidence of virtual-to-real traversal [TIRTA Ventures, retrieved 2026].
Pricing uses tiered plans for studios of varying size. Enterprise offers a customer success manager, 24/7 support, priority resolution, and onboarding [nunu.ai, retrieved 2024]. Dollar pricing and customer counts beyond StormForge are not public.
Data Accuracy: GREEN -- Confirmed by nunu.ai product pages, Y Combinator, and FOV Ventures.
Market Research and Opportunity
PUBLIC
Game QA has resisted automation despite being labor-heavy. Agent-based testing now draws investor attention.
AAA titles need hundreds of testers across platforms and locales. Live-service games make QA continuous.
nunu.ai sells into this with validation from gaming investors a16z speedrun and TIRTA Ventures [Crunchbase, March 2025] [StartupTicker].
A precise TAM for AI game QA is not public from third-party reports.
a16z speedrun backs games infrastructure. TIRTA framed nunu.ai around QA automation budgets [TIRTA Ventures, retrieved 2026]. FOV Ventures bets on agents generalizing across environments [FOV Ventures, retrieved 2026].
Three demand vectors matter.
First, mobile and live-service publishers face heavy QA. nunu.ai's web model and input layer align.
Second, agent playthroughs could feed marketing analytics.
Third, embodied AI and robotics simulation is adjacent. "Rob the Robot" gestures at it [FOV Ventures, retrieved 2026].
Regulatory exposure is low. Cost discipline since 2023 pressures budgets (automation tailwind) but slows vendor adoption.
| Signal | Value | Source |
|---|---|---|
| Seed round size | $6,000,000 | [Crunchbase, March 2025] |
| Total disclosed funding | ~$8,000,000 | [StartupTicker] |
| Disclosed reference customer | StormForge (case study) | [nunu.ai, retrieved 2024] |
Specialist investors believe AI QA is fundable. Market-size lacks a named report, so sizing is directional.
Data Accuracy: YELLOW -- Investor framing and round data are confirmed; a named TAM report is not in the cited evidence.
Competitive Landscape
MIXED
nunu.ai is a pure-play, engine-agnostic AI QA agent. This sets it apart from QA services and in-house tools. Research names no direct competitors [PUBLIC].
Competition has three layers.
Incumbents are outsourcing firms like Keywords Studios, PTW, Testronic. They sell human tester hours with procurement ties and localization depth. Labor cost exposes them to substitution.
In-house tools at large studios couple to engines or franchises. Publishers like EA and Activision have playtest bots. nunu.ai claims to remove that constraint [Y Combinator].
AI-native QA startups emerge in YC and agent categories. Request a competitive map from the company.
nunu.ai's edge: demos like Hogwarts Legacy and Pokémon Emerald show long-horizon tasks without scripting [nunu.ai, retrieved 2024]. The nexus portal aids non-engineers.
The edge perishes as multimodal models improve. Moat shifts to product surface, test library, customer data [PUBLIC].
Exposure: large publishers build internally on frontier models, especially for IP-sensitive franchises.
In 18 months, nunu.ai wins with mid-to-large studio logos and hours-saved metrics. It loses if incumbents bundle AI agents into service agreements.
Data Accuracy: YELLOW -- Competitor identities are inferred from category knowledge; no direct competitor was named in the cited research.
Opportunity
PUBLIC
If the agent generalizes as demos suggest, nunu.ai could become the default QA automation layer. It might grow larger.
The headline opportunity. nunu.ai could define QA for interactive entertainment, like test vendors in web/mobile.
Evidence: Hogwarts Legacy demo [nunu.ai, retrieved 2024]. a16z speedrun lead offers distribution [Crunchbase, March 2025]. StormForge case study shows workflow integration [nunu.ai, retrieved 2024].
Growth scenarios.
| Scenario | What happens | Catalyst | Why it's plausible |
|---|---|---|---|
| AAA standardization | nunu.ai becomes a procurement-line item across two or more top-10 publishers | A reference deployment at an a16z speedrun portfolio studio that publishes hours-saved metrics | a16z speedrun's gaming-specific mandate gives nunu.ai a warm channel into studio leadership [Crunchbase, March 2025] |
| Mobile and live-service wedge | The product becomes the default continuous-QA layer for live-service titles where build cadence is weekly | Integration with mobile build pipelines and a published case beyond StormForge | Live-service economics make per-build human QA increasingly uneconomic, and nexus already files into Slack and Jira [nunu.ai, retrieved 2024] |
| Embodied agents pivot-and-add | The agent stack becomes a dual-use platform serving both game QA and physical robotics simulation | A funded research partnership extending the "Rob the Robot" experiment | Investor framing already names virtual-to-physical generalization as part of the thesis [TIRTA Ventures, retrieved 2026] [FOV Ventures, retrieved 2026] |
What compounding looks like. Customer tests produce inputs, traces, reports that improve agents [nunu.ai, retrieved 2024]. Test plans create switching costs at QA leadership.
Distribution: a16z referrals, YC network.
The size of the win. No direct AI QA peer. Keywords Studios went private in multi-billion-pound deal, sizing the pool [2024]. Software margins beat services.
Data Accuracy: YELLOW -- Capability claims and investor framing are sourced; scenarios and comparable valuations are explicitly labelled directional.
Sources
PUBLIC
[nunu.ai] Home | https://nunu.ai/
[nunu.ai] About | https://nunu.ai/about
[nunu.ai] Pricing | https://nunu.ai/pricing
[nunu.ai] Enterprise | https://nunu.ai/enterprise
[nunu.ai] Case Studies | https://nunu.ai/case-studies
[nunu.ai] StormForge case study | https://nunu.ai/case-studies/stormforge
[nunu.ai] Bridging the Virtual and Physical World with nunu.ai (Rob the Robot) | https://nunu.ai/blog/rob-the-robot
[nunu.ai] Beating the World Record in Pokémon AI Live on Stream | https://nunu.ai/blog/ai-pokemon
[nunu.ai] Trust Center | https://trust.nunu.ai/
[Y Combinator] nunu.ai company profile | https://www.ycombinator.com/companies/nunu-ai
[Crunchbase] nunu.ai company profile | https://www.crunchbase.com/organization/nunu-ai
[Crunchbase, March 2025] Seed Round - nunu.ai - 2025-03-12 | https://www.crunchbase.com/funding_round/nunu-ai-seed--913d80c1
[Crunchbase] Pre-Seed Round - nunu.ai | https://www.crunchbase.com/funding_round/nunu-ai-pre-seed--eecb1fc8
[StartupTicker] Nunu.ai closes $6 Million seed round | https://www.startupticker.ch/en/news/nunu-ai-closes-6-million-seed-round
[Tracxn] Nunu.ai 2025 Company Profile, Team, Funding & Competitors | https://tracxn.com/d/companies/nunu.ai/__PbrNxM4dOVCEHrLMZwfQaBcsEgAn09l5LqcmDt75B9w
[LinkedIn] nunu.ai (YC W23) company page | https://www.linkedin.com/company/nunu-ai
[FOV Ventures, retrieved 2026] Investing in Nunu.ai | https://viewpoints.fov.ventures/p/investing-in-nunu-ai
[FOV Ventures, retrieved 2026] nunu.ai job board | https://jobs.fov.ventures/companies/nunu-ai
[TIRTA Ventures, retrieved 2026] Nunu.ai raises $6M for AI agents dubbed 'unembodied minds' for game testing | https://www.tirta.io/post/nunu-ai-raises-6m-for-ai-agents-dubbed-unembodied-minds-for-game-testing
[PrivCo, retrieved 2026] nunu.ai company profile | https://system.privco.com/company/nunuai
[AInvest] nunu.ai Secures $6M Seed Round, Led by TIRTA Ventures and a16z Speedrun | https://www.ainvest.com/news/nunu-ai-secures-6m-seed-led-tirta-ventures-a16z-speedrun-2503/
[PitchBook] nunu.ai 2026 Company Profile | https://pitchbook.com/profiles/company/589194-37
Articles about nunu.ai
- nunu.ai Wants an AI Agent Playing Through Every Boss Fight Before Your QA Team Wakes Up — The San Francisco startup raised $6M from a16z speedrun to point multimodal agents at the most tedious job in game development.