What daisytuner does

Daisytuner GmbH, founded in 2024 and headquartered in Darmstadt, Germany, builds an AI-accelerated, self-learning compiler stack that compiles applications written in PyTorch, NumPy, and C/C++ into optimized native code for CPUs, GPUs, and emerging accelerators including RISC-V (Tenstorrent Blackhole) and photonic processors (Q.ANT NPU Gen 2). The core technology centers on Stateful Dataflow Multigraphs (SDFGs) as a cross-framework intermediate representation, the Daisytuner Optimizing Compiler Collection (docc) — a drop-in replacement for clang/gcc with a Python @native decorator — and a cloud-connected optimization database that continuously ingests performance profiles to improve compilation. Products include Daisyflow (dataflow graph capture), docc compilers for C/C++ and Python, Daisy Cloud (a multi-architecture CI/CD platform distributed as a GitHub App), self-hosted runners, and a Daisy Dashboard for monitoring. The company sells via a hybrid PLG motion (pip install, GitHub App, public open-source under BSD-3) plus a 'Book a Demo' enterprise motion, with revenue from usage-based cloud compute and subscription/enterprise self-hosted licenses; no public pricing is disclosed. Daisytuner's primary customers are enterprise HPC and simulation operators (manufacturing, aerospace, automotive, energy, research), AI/ML engineers running models across heterogeneous hardware, and emerging chipmakers seeking to broaden their software ecosystem. Strategic partners include ETH Zurich's SPCL lab (technology transfer origin), Tenstorrent, Q.ANT, and TU Darmstadt.

Name

daisytuner

Legal name

Daisytuner GmbH

Website

https://daisytuner.com

Company type

Private

Founded year

2024

Operating status

Operating

Headcount range

1–10 employees

Short description

Daisytuner builds an AI-guided, self-learning compiler that compiles PyTorch, NumPy, and C/C++ into optimized native code for CPUs, GPUs, RISC-V, and photonic processors, serving HPC simulation operators, AI/ML engineers, and emerging chipmakers through a cloud platform and self-hosted enterprise licenses.

Ownership category

akta.pro rank

Product category

Compiler Optimization Software

NAICS

Custom Computer Programming Services (541511), Computer Systems Design and Related Services (5415)

SIC

Services-Computer Programming Services (7371), Services-Prepackaged Software (7372)

akta.pro primary industry

End-to-End MLOps & ML Platform Suites (HDAAABAA)

akta.pro secondary industry

Edge AI Model Optimization & Compression (quantization, pruning, distillation) (HDAAAJAA)

Keywords

• Compiler optimization software
• AI code compilation
• Heterogeneous hardware compilation
• Performance portability tools
• Developer productivity tools

Headquarters

HQ city

Darmstadt

HQ country

Germany

HQ region

Europe

Offices1 record

Markets served

GTM type

B2B

Offering type

Software

Cost components

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

Revenue model

1. Daisy Cloud - Cloud Compute Access: Daisytuner offers cloud-based access to heterogeneous compute clusters (AMD, NVIDIA, Tenstorrent, Q.ANT photonic, ARM) via its Daisy Cloud platform. Users run workflows on cloud runners and pay for compute time. The platform integrates as a GitHub App for CI/CD use.
2. docc Compiler - Self-Hosted / Enterprise Licenses: The docc compiler can be installed and self-hosted (self-hosted runners). The platform supports enterprise configurations with custom partitions and dedicated hardware, likely sold as subscription or seat-based licenses.

Go-to-market motion2 records

Distribution channels5 records

Marketing channels6 records

Core offering

Daisytuner provides an AI-accelerated, self-learning compiler stack (docc) that compiles applications written in PyTorch, NumPy, and C/C++ into optimized native code for CPUs, GPUs, and novel accelerators (photonic processors, RISC-V) without code rewrites. Its Daisyflow technology captures multi-framework applications as unified dataflow graphs (Stateful Dataflow Multigraphs / SDFGs) for end-to-end optimization. The platform includes Daisy Cloud for CI/CD benchmarking and self-hosted runner infrastructure accessible via a GitHub App integration.

Product overview

Daisytuner provides an optimization platform for modern compute that enables software to run across CPUs, GPUs, and accelerators without rewriting code. The core products are Daisyflow (dataflow graph capture technology), docc (Daisytuner Optimizing Compiler Collection), and Daisy Cloud (cloud benchmarking and deployment platform). The platform uses a self-learning compiler that leverages neural networks and an optimization database to automatically optimize applications for different hardware targets. Daisy Cloud provides CI runners and self-hosting options, while the compiler supports PyTorch, NumPy, and C/C++ frameworks targeting NVIDIA, AMD, ARM, Q.ANT photonic processors, and Tenstorrent accelerators.

Differentiator

Problem solved

Functional benefit

Products and services

• Daisyflow Dataflow graph capture technology that sees across framework boundaries, allowing multi-framework applications (PyTorch, NumPy, C/C++) to be captured as a single dataflow graph for unified end-to-end optimization.
• docc (Daisytuner Optimizing Compiler Collection) Self-learning compiler that uses neural networks to generate code embeddings and queries a cloud-connected optimization database. Acts as a drop-in replacement for clang and gcc that compiles PyTorch, NumPy, and C/C++ code for CPUs, GPUs, and accelerators (NVIDIA, AMD, ARM, Tenstorrent RISC-V, Q.ANT photonic) with end-to-end optimization.
• Daisy Cloud Cloud backend providing access to CI runners with multiple processor types (AMD, NVIDIA, Tenstorrent RISC-V, Q.ANT photonic, ARM) for multi-architecture building, benchmarking, and testing in CI/CD pipelines. Distributed as a GitHub App via the GitHub Marketplace and supports workflow definitions in .daisy/ YAML files.
• Self-Hosted Runners On-premises runner infrastructure similar to GitHub runners, enabling organizations to execute tests, performance evaluation, and autotuning workloads on their own hardware with detailed performance metrics capture. Uses Docker containers (Docker CE required) to run workflow steps and benchmarking workloads, supporting enterprise configurations with custom partitions and dedicated hardware.
• Daisy Dashboard Web-based dashboard for monitoring compilation progress, inspecting offloaded code parts, viewing performance metrics, and managing partitions and runners on the Daisy Cloud platform.

Quantifiable outcome

• 10% faster than highly-tuned Fortran code for CLOUDSC cloud microphysics scheme
• +3 more outcomes

Segments3 records

Ideal customer profiles3 records

Technology focussed Yes

API detail

Has API

No

API docs

API detail

Core technology

AI maturity

App detail

Integration2 records

AI capability4 records

Feature6 records

Partnerships

Four partnerships are on record, tiered core, flagship and minor.

• ETH Zurich / SPCL (Scalable Parallel Computing Laboratory)coreStrategic or Co-development PartnerDaisytuner was founded on technology transferred from ETH Zurich's SPCL lab. The partnership involves the Data-Centric Programming (DaCe) framework transfer and ongoing joint research. SPCL, led by Prof. Torsten Hoefler (multiple ACM Gordon Bell Prize winner), is one of the world's leading HPC laboratories. The companies collaborate on academic publications (CGO 2025 paper with Lukas Truemper) and have won the HiPEAC Technology Transfer Award together.
• TenstorrentcoreTechnology or IntegrationDaisytuner collaborates with Tenstorrent to cross-compile industry-standard HPC applications (OpenFOAM, HPCCG) to Tenstorrent's RISC-V based Blackhole accelerator. Daisytuner wrote optimized BLAS routines (sparse matrix-vector multiplication) for Tenstorrent hardware. OpenFOAM-TT and HPCCG projects are publicly available on GitHub. The collaboration enables Tenstorrent to address markets beyond AI.
• Q.ANTflagshipTechnology or IntegrationDaisytuner achieved a world-first by compiling and deploying a standard PyTorch object detection model (Faster R-CNN with ResNet-50) directly onto Q.ANT's photonic NPU Gen 2 — the first time a standard ML framework targeted photonic hardware. Full pipeline (pre-processing, inference, post-processing) runs end-to-end with no custom code required from the developer.
• TU DarmstadtminorStrategic or Co-development PartnerDaisytuner team members (including Dr. Nora Hagmeyer) lecture at TU Darmstadt's Industry Colloquium on compilers for novel processors, hosted by Prof. Dr.-Ing. Christian Hochberger's Institute for Data Technology. Represents a talent pipeline and research collaboration.

Scale indicators3 records

Recent moves6 records

Expansion highlights6 records

Direct peers

• OctoML (OctoAI): Pioneer in ML model optimization and deployment across heterogeneous hardware using Apache TVM. Directly comparable as a compiler/runtime optimization platform targeting multiple accelerators from a single model artifact, addressing the same hardware portability pain point.
• Modular: Builds the Mojo language and AI inference platform aimed at unifying AI compute across hardware. Closely comparable as a compiler-first stack for portable AI workloads with developer-tools-led adoption, competing for the same ML portability customer base.

Broad incumbents

• SambaNova Systems: Builds custom AI silicon (RDU) alongside a proprietary compiler stack to optimize models for its hardware. Comparable as a vertically-integrated chip-plus-compiler play targeting enterprise AI performance, though with proprietary rather than portable hardware.
• Cerebras Systems: Wafer-scale AI accelerator company with its own compiler/runtime stack (Cerebras Software Platform). Comparable as a chip-plus-software stack addressing performance portability, though tied to its proprietary silicon rather than enabling cross-vendor portability.
• Groq: LPU-based AI inference company with a custom compiler stack that compiles models from standard frameworks onto its hardware. Comparable as a vertically-integrated platform addressing the same hardware lock-in problem Daisytuner solves, though for a single proprietary target.
• Graphcore: IPU-based AI compute company with Poplar software stack that compiles standard ML frameworks to its custom hardware. Comparable as a chip-plus-compiler vendor serving the same HPC/AI customer segments and tackling the same portability challenge.

Emerging players

• Tenstorrent: RISC-V AI accelerator company that is both a strategic partner and potential competitor; Tenstorrent maintains its own compiler stack (Buda/TT-Metalium) for its silicon. Overlaps with Daisytuner on Tenstorrent hardware optimization but competes on full-stack ownership of the compiler.
• Codon: Open-source Python-to-native compiler for high-performance numerical computing. Comparable in the niche of Python-to-native compilation for HPC and AI workloads, though without Daisytuner's multi-hardware accelerator targeting or self-learning optimization database.

Others

• Anyscale: Ray-based distributed compute platform enabling scalable AI workloads across heterogeneous clusters. Adjacent as a developer-platform layer for portable AI workloads, though focused on distributed orchestration rather than compiler-level hardware optimization.
• Codeplay Software: Specialist in SYCL and DPC++ compiler implementations for heterogeneous hardware including accelerators. Adjacent infrastructure peer providing parallel-computing compiler expertise that could integrate with or compete against parts of Daisytuner's stack.

Market position

Strengths5 records

Weaknesses5 records

Competitive moat5 records

Key risks6 records

Key highlights7 records

Customer concentration

• GitHubgithub.com/daisytuner
• LinkedInlinkedin.com/company/daisytuner

Compliance2 records

Revenue estimate

Valuation estimate

Number of profiles

Profiles5 records

Funding overview

Funding rounds

Investors

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 daisytuner