What Myrtle Software does

Myrtle Software Ltd, operating commercially as myrtle.ai, is a Cambridge, UK-based AI infrastructure company that builds FPGA-based machine learning inference accelerators. Its core product, VOLLO, is a software stack comprising the Vollo Compiler (a PyTorch/ONNX frontend), the Vollo Runtime (C and Python APIs), pre-built FPGA bitstreams, and an IP core for custom designs. VOLLO targets workloads where inference latency is the binding constraint, most notably capital-markets trading, and achieves audited latencies of 2 microseconds at the 99th percentile on STAC-ML benchmarks. Adjacent brands CAIMAN (real-time speech / conversational AI) and SEAL (recommendation) extend the same inference IP into new verticals, alongside vertical variants of VOLLO for telecoms and security.

The technology differentiates on deterministic, microsecond-scale latency and high compute density (up to 10x per server in 1U form factor), made possible by co-designed FPGA bitstreams running on AMD Versal Premium and Intel Agilex accelerator cards sourced from partners such as Silicom, BittWare, Napatech, and Supermicro. Software is consumed via free evaluation (the Vollo Sandbox) or under per-device token licensing tied to specific accelerator device IDs; a separate Vollo IP Core licensing path supports integration into customers' own FPGA designs. Distribution is hybrid: product-led self-serve evaluation plus direct enterprise sales, with dedicated fintech sales coverage.

Commercial traction is concentrated in capital markets — hundreds of thousands of hours of production trading on VOLLO at leading global trading firms — while the newer CAIMAN, SEAL, telecoms, and security lines remain early-stage. The company is privately held by its founder/CEO Peter Baldwin with an executive chairman and director, has 11–50 employees, and has historically been backed by Cambridge Angels, Delin Ventures, IQ Capital, and Origin Capital.

Name

Myrtle Software

Legal name

Myrtle Software Ltd

Website

https://myrtle.ai

Company type

Private

Founded year

2003

Operating status

Operating

Headcount range

11–50 employees

Short description

Myrtle Software (trading as myrtle.ai) builds FPGA-based AI inference accelerators delivering microsecond-latency machine learning predictions, primarily for capital-markets trading firms, with secondary applications in telecoms, security, conversational AI, and recommendation systems.

Ownership category

akta.pro rank

Product category

AI Inference Acceleration Software

NAICS

Software Publishers (513210)

SIC

Services-Computer Programming Services (7371)

akta.pro primary industry

AI/Neural Processing Units (NPUs) & ML Accelerators (HDAHAJAF)

akta.pro secondary industries

Compute & Acceleration ICs (CPU/GPU/AI/FPGA/ASIC) (HDAHABAC), On-Device Inference Runtimes & SDKs (mobile/embedded) (HDAAAJAB)

Keywords

• FPGA inference acceleration
• Low latency machine learning
• AI inference accelerator
• Capital markets AI
• Vollo software stack

Headquarters

HQ city

Cambridge

HQ country

United Kingdom

HQ region

Europe

Offices1 record

Markets served

GTM type

B2B

Offering type

Software

Cost components

Personnel, Technology or R&D, Marketing or Sales, Operations, Infrastructure

Revenue model

1. Software Licensing: Per-device licensing model where customers purchase tokens to redeem licenses for specific accelerator devices. Licenses are tied to device IDs and enable operation of Vollo software on FPGA hardware.
2. Professional Services: Implementation and support services likely offered alongside software licensing for enterprise customers deploying VOLLO in production environments.

Pricing tiers

Go-to-market motion2 records

Distribution channels3 records

Marketing channels6 records

Core offering

Myrtle Software builds and licenses VOLLO, an FPGA-based AI inference accelerator that runs machine learning models in microseconds (2 microseconds at the 99th percentile in STAC-ML benchmarks). The offering is sold as per-device software licenses (Vollo SDK, Compiler, Runtime, IP Core) targeting capital markets trading, telecoms, security, conversational AI, and recommendation system applications, and runs on AMD Versal Premium and Intel Agilex FPGA accelerator cards supplied by hardware partners.

Product overview

Myrtle Software (operating as myrtle.ai) offers a portfolio of low-latency AI inference accelerators delivered through two product lines: VOLLO (with vertical variants for capital markets, telecoms, and security) and CAIMAN (for conversational AI/speech processing), plus SEAL (for recommendation). The core technology is the Vollo SDK, which provides compiler tools to transform PyTorch/ONNX models into optimized programs for FPGA-based hardware accelerators. The company also offers Vollo IP Core for customers wanting to integrate the technology into custom FPGA designs. VOLLO targets applications requiring microsecond-level inference latency, with the capital markets variant achieving 2 microseconds at the 99th percentile in STAC-ML benchmarks.

Differentiator

Problem solved

Functional benefit

Brands

• VOLLO: Low latency inference accelerator for capital markets, telecoms, and security applications.
• CAIMAN
• SEAL

Products and services

• VOLLO Flagship low-latency AI inference accelerator software for FPGA-based platforms; delivers microsecond-level ML inference (as low as 2 microseconds at the 99th percentile) and supports MLP, CNN, and LSTM models compiled from PyTorch or ONNX. Targeted at enterprise customers in capital markets, telecoms, and security who need microsecond ML inference.
• VOLLO for Capital Markets VOLLO configured for financial trading applications, enabling microsecond AI inference for trading decisions, risk analysis, and quote generation. Targets trading firms requiring ultra-low latency for competitive advantage.
• VOLLO for Telecoms VOLLO configured for wireless telecoms applications, providing low-latency inference for real-time decision-making in network operations and optimization. Targets telecoms operators running ML on streaming network data.
• VOLLO for Security VOLLO configured for security applications, providing faster threat detection capabilities for network protection. Targets enterprise security teams and fraud detection platforms requiring microsecond anomaly detection.
• CAIMAN CAIMAN is Myrtle's real-time speech processing, transcription and analysis product for conversational AI applications. Targets developers of conversational AI platforms requiring low-latency speech inference.
• SEAL SEAL is Myrtle's recommendation engine accelerator product that improves recommendation relevance with faster throughput for personalization applications. Targets e-commerce and content platforms with high-throughput recommendation needs.
• Vollo SDK Software development kit enabling low-latency streaming inference of ML models on FPGA platforms. Includes the vollo-torch PyTorch frontend, vollo-compiler backend, Vollo Runtime (C API and Python bindings), and pre-built FPGA bitstreams for supported accelerator cards. Supports Python 3.7+.
• Vollo Compiler Compiler that transforms PyTorch or ONNX neural network models into optimized .vollo program files for execution on Vollo accelerators; includes a virtual machine for performance estimation without hardware. Supports MLP, CNN, LSTM models and ONNX import.
• Vollo Runtime Low-latency asynchronous inference API for running compiled Vollo programs on FPGA accelerators; available as a C API and Python bindings, supporting bfloat16 and float32 data formats with DMA optimization.
• Vollo IP Core Encrypted netlist IP core that can be integrated into customers' own FPGA designs. Available upon request for Agilex FPGAs, scalable by number of cores and core size.
• Vollo Sandbox Online evaluation environment allowing ML developers to test model performance on Vollo without needing FPGA hardware or a Vollo license.

Quantifiable outcome

• 2 microseconds latency at 99th percentile (STAC-ML audited)
• +4 more outcomes

Named customers2 records

Segments5 records

Ideal customer profiles4 records

Technology focussed Yes

API detail

Has API

Yes

API docs

API detail

Core technology

AI maturity

App detail

AI capability5 records

Feature8 records

Partnerships

Four partnerships are on record, tiered core.

• AMDcoreTechnology or IntegrationAMD is a core technology partner providing Versal Premium series VP1802 Adaptive SoC for VOLLO inference acceleration. The AMD Versal Premium Series provides PCIe Gen5x8 and more than 3.3M programmable LUTs, enabling ultra-low latency inference. Partnership includes STAC-ML benchmark collaboration.
• Silicom DenmarkcoreTechnology or IntegrationSilicom provides FPGA accelerator cards including the FBAP4@VP18-2L0S PCIe accelerator card and the Artena accelerator card featuring AMD Versal Premium FPGAs. These cards are the hardware platform for VOLLO in STAC-ML benchmarks and production deployments.
• SupermicrocoreTechnology or IntegrationSupermicro provides server infrastructure (AS-2015CS-TNR) used in STAC-ML benchmark testing of VOLLO. Supermicro's AMD systems address challenging workloads in financial services with top-end performance and very low latencies.
• BittWarecoreTechnology or IntegrationBittWare provides IA-420F and IA-840F FPGA accelerator cards based on Intel Agilex FPGAs that support the VOLLO software stack. These are officially supported accelerator cards in the Vollo SDK.

Scale indicators5 records

Recent moves6 records

Expansion highlights5 records

Direct peers

• Mipsology: Mipsology provides FPGA-based AI inference acceleration (Zebra software stack) targeting very low-latency edge and data-center inference - same fundamental approach (FPGA mapping of neural networks) and similar buyers (enterprises needing low-latency AI) as Myrtle's VOLLO.

Emerging players

• Groq: Groq builds low-latency LPU inference accelerators targeting deterministic, microsecond-class latency for ML inference - directly comparable technical thesis to Myrtle's VOLLO, though implemented in custom ASIC silicon rather than FPGA.
• SambaNova Systems: SambaNova offers a reconfigurable dataflow AI accelerator platform for enterprise inference, including financial-services workloads - overlapping with VOLLO's positioning for low-latency ML inference, but using bespoke RDU silicon rather than commodity FPGAs.
• Cerebras Systems: Cerebras provides wafer-scale AI compute for high-throughput inference and training, including financial services applications - serving the same enterprise inference buyer seeking speed advantages as Myrtle's flagship VOLLO offering.
• SiMa.ai: SiMa.ai builds a Machine Learning SoC (MLSoC) for low-power, low-latency edge AI inference on an embedded platform - similar deterministic, microsecond-inference thesis and product philosophy as VOLLO's on-FPGA acceleration, with CAIMAN/SEAL verticals as closest functional analogs.

Broad incumbents

• NVIDIA: NVIDIA is the dominant GPU-based AI inference provider and defines the alternative architecture Myrtle's VOLLO is positioned against - they sell Triton/ TensorRT software stacks and DGX/HGX hardware covering trading, telecoms, security and conversational AI workloads across the same buyer base.
• Habana Labs (Intel): Habana Labs, owned by Intel, provides Gaudi/Habana accelerators and SynapseAI software for data-center training and inference - a competing broad-portfolio AI acceleration incumbent with enterprise software stacks, often benchmarked against VOLLO in the inference category.

Others

• OctoML: OctoML provides automated MLOps and inference optimization (Apache TVM-based) for diverse hardware backends - functionally complementary to Myrtle's Vollo Compiler/Runtime path, easing the friction of porting PyTorch/ONNX models onto any accelerator including FPGAs.

Market position

Strengths4 records

Weaknesses4 records

Competitive moat5 records

Key risks5 records

Key highlights6 records

Customer concentration

• Xx.com/Myrtle_ai
• LinkedInlinkedin.com/company/myrtle-software

Revenue estimate

Valuation estimate

Number of profiles

Profiles3 records

Funding overview

Funding rounds3 records

Investors4 records

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about Myrtle Software