“LGTM, thanks for the PR”
leslie-fang25 · NVIDIA reviewer — merged it
+40%throughput
mergedJun 2026 · PR #15550
TensorRT-LLM · NVIDIAGet more from systems you already run
Improve performance across the models, agents, and code you already run while reducing infrastructure costs. Artemis explores multiple optimization strategies in parallel, tests them on your workload, and delivers only what is measurably better.
Animated demo: a user asks Artemis to make their code faster and use less memory without changing what it returns. Artemis scans the file, fixes a small issue with a coding agent, hands the slow section to evolutionary search, benchmarks candidates on a latency versus memory trade-off, picks a winner, and opens a verified pull request.
process.cpppipelinedecoder.cppmodelsutils.hppshared
idle01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
Waiting for a goal.
- Read process.cpp
- Fix ISS-9 with a coding agent
- Locate the hot section
- Hand ISS-6 to evolution
- Benchmark candidates against the goal
- Open a verified pull request
Pull RequestCreate a PR
process.cpp
+
-
+
+
+
Evolutionreceiving…
cross-attn424 µs4.9×RTF0.1349−16%
Lloyds Bank
intel.
||Deaglo
CBCredit
Benchmark
Benchmark
BOSTON
Open Platform
for Enterprise AI
for Enterprise AI
Exasol
TaylorWessing
Hazelcast
|1NayaOne
Lloyds Bank
intel.
||Deaglo
CBCredit
Benchmark
Benchmark
BOSTON
Open Platform
for Enterprise AI
for Enterprise AI
Exasol
TaylorWessing
Hazelcast
|1NayaOne
Lloyds Bank
intel.
||Deaglo
CBCredit
Benchmark
Benchmark
BOSTON
Open Platform
for Enterprise AI
for Enterprise AI
Exasol
TaylorWessing
Hazelcast
|1NayaOne
Lloyds Bank
intel.
||Deaglo
CBCredit
Benchmark
Benchmark
BOSTON
Open Platform
for Enterprise AI
for Enterprise AI
Exasol
TaylorWessing
Hazelcast
|1NayaOne
Built for every optimization challenge
From AI inference to data pipelines, explore how Artemis optimizes what's already in production.
Serve more users on the hardware you already own
Throughput, tokens per second
Baseline0 t/s
Artemis0 t/s
2.07×
faster inference
vLLM, Qwen 3.6 35B, Intel Xeon 6. Outputs unchanged.
Changes engineers were happy to merge.
Explore pull requests proposed by Artemis and merged by the maintainers of widely used open source projects.
- TensorRT-LLM · NVIDIA
4.9× on the attention call — A permute left the tensor non-contiguous, so PyTorch quietly dropped off the fused attention path onto the slow one.
jongwook · A Whisper author — merged it
4.9×on the attention callmergedJul 2026 · PR #2812
Whisper · OpenAI5.2× and 47× on the copy kernels — The quantized copy kernels ran one thread per block, so 31 of every 32 warp lanes sat idle, and around 97% of blocks exited immediately at the bounds check.
ggerganov · Created llama.cpp — merged it
5.2× & 47×on the copy kernelsmergedAug 2026 · PR #26731
llama.cpp
Teaching software to improve itself.
TurinTech began long before today's wave of generative AI.
As a UCL spinout, our team spent years researching how software could continuously improve through evolutionary optimization.
Today, Artemis combines that research with generative AI to continuously improve code, AI models, agents and infrastructure at enterprise scale.
Deploy on your terms.
Run Artemis where your code already lives, from your own laptop to a fully hosted deployment. Vendor and model neutral, built to fit your existing stack.
deploymentBring your key
Run it locally
one model or many
artemis.cloud
discovery runlive
EXP-9DF86%
EXP-B1074%
v1.0★ best92%
deployed anywhere
Anthropicartemis.cloudvpc-eu-west-1
your keys, your boundary
Ready to route
Fresh from the team.
Common questions.
If you can measure it, Artemis can find a better version and prove it. In practice that means serving more users on the hardware you own, spending less token per agent task, and shipping models and code that perform better than the ones you have. We’ve optimized AI inference, AI agents, latency-critical systems, data and SQL, decision algorithms, and ML models.
Discover the ROI hiding in your stack
Point Artemis at a system you already run, and see the improvement it finds, validated, before you change a thing.