Pomegra
Guava, a Los Angeles voice-agent startup, raised a $12.5M seed led by Acrew Capital to scale Daytona, its in-house voice AI model for regulated industries.
Key Takeaways
- Guava closed a $12.5M seed round led by Acrew Capital; the valuation was not disclosed.
- Its Daytona model runs speech recognition, language and speech synthesis as one stack.
- Daytona launched in September and already serves more than a dozen enterprise customers.
Lead
Guava, a Los Angeles developer platform for AI voice agents, announced a $12.5 million seed round on November 3, 2026. Acrew Capital led the financing. Flourish Ventures, WillowTree Ventures, Clocktower Ventures and Commerce Ventures, which backed Guava at pre-seed, also participated. The company will spend the money on model and product development and on a go-to-market team aimed at enterprises and business process outsourcers (BPOs). Valuation and total capital raised were not disclosed.
What Does Guava Actually Build?
Guava builds a developer platform for voice agents that handle customer calls such as appointment scheduling and live support. Most competing platforms stitch together speech recognition, a language model and speech synthesis from separate vendors. Guava trains and runs its own models across all three layers.
Its first public model, Daytona, launched in September 2026. Developers can build a working agent in under five minutes on a free tier, using Python, TypeScript or Claude Code. Enterprise customers typically put their first live regulated call into production within 14 days. Guava says Daytona ranks in the top 10 on Coval's word error rate benchmark.
Who Is Behind the Company?
The founding team has roots in voice AI and in payments. Omri Dahan, the CEO, spent more than a decade at Marqeta, where he served as chief revenue officer, and previously invested in Gridspace. Anthony Scodary, co-head of engineering, and Nico Benitez are technical co-founders from Gridspace, a voice AI company. Scodary has more than 10 years of experience in the field. Mike Piechna is the fourth co-founder.
That combination matters for the pitch. The technical founders have built speech systems before, and the CEO has sold into regulated financial buyers.
Why Does Guava Focus on Regulated Industries?
Guava targets regulated industries such as insurance because voice agents that perform well in demos tend to fail on live calls, a gap Dahan has described directly. "We built Guava to give enterprises agents they can trust with high-volume, high-stakes conversations," he said in the announcement. In regulated settings a misheard policy number or a misstated disclosure carries compliance consequences, not just an annoyed caller.
Controlling the full model stack is Guava's answer. A single vendor owns the failure modes, from transcription errors to latency, rather than pointing at a third-party component. The tradeoff is cost: training and serving proprietary models takes more capital and engineering time than assembling APIs, which a $12.5M seed will not stretch far.
How Crowded Is the Voice Agent Market?
The market is crowded, and Guava is entering with a narrow thesis. Market forecasts cited in the announcement put voice agent revenue at $5.1 billion in 2026, rising to $43.8 billion by 2036. Such projections are easy to publish and hard to verify, so the more useful signal is the number of well-funded startups competing for the same enterprise call-center budgets.
Guava's current traction is modest but concrete. More than a dozen enterprise customers run Daytona, which handles tens of thousands of calls per day. At that volume the company is past the prototype stage, though revenue, retention and pricing remain undisclosed. A seed round of this size, with no disclosed valuation, suggests investors are paying for the team and the model approach before the commercial record exists.
What Comes Next?
The company's stated roadmap points to edge-device deployments, on-premises installations and full-duplex interaction, where an agent can listen and speak at the same time as a human does. On-premises options matter most for banks and insurers that restrict where call audio can travel. Full-duplex conversation is harder, and it is where many voice agents still sound stilted.
Hiring will be the first test. Selling to BPOs and large enterprises requires a sales and implementation team that the founders have not yet built at scale. The second test is whether benchmark standing and a 14-day deployment time hold up as call volumes grow by an order of magnitude.
Outlook
Guava has $12.5M, a lead investor in Acrew Capital, and a product already handling tens of thousands of daily calls for regulated customers. Its bet is that owning the full voice stack beats assembling third-party parts when errors carry legal weight. Whether that holds will depend on enterprise conversion, the cost of running proprietary models, and how quickly larger rivals close the reliability gap.
