LOCI is an AI‑driven observability platform that analyzes compiled CPU and GPU binaries, using a hardware‑aware large code language model to predict performance and power hotspots before test or inference runs. It automatically rewrites binaries and adjusts runtime configurations, integrating with CI/CD pipelines to provide measurable throughput and energy savings for AI/ML and performance engineering teams.
Funding
$63M raised to dateRaised to date based on public sources. This may differ from the amount the company actually raised and is based only on what is publicly available on the internet.
MSFounders
Product
Problem
Engineering teams spend a large portion of their time diagnosing performance regressions, power spikes, and bottlenecks in compiled binaries, especially for AI/ML workloads. Manual tuning is error‑prone and often leads to over‑provisioned GPU/CPU resources, inflating compute costs and reducing overall system efficiency. The lack of early‑stage observability makes it difficult to predict issues before they reach production.
Solution
LOCI is an AI‑driven observability and performance‑intelligence platform that operates directly on compiled BIN files. By leveraging a hardware‑aware Large Code Language Model (LCLM), LOCI predicts hotspots, thermal spikes, and throughput inefficiencies before test or inference runs. The platform then autonomously applies opcode‑level optimizations to code, runtime configurations, and serving pipelines, delivering higher throughput per watt. Integrated with CI/CD pipelines, LOCI provides continuous feedback and evidence‑backed reports that quantify power savings and performance gains. The solution reduces the need for manual debugging, curtails over‑provisioning, and enables engineering teams to meet SLA targets with lower infrastructure spend.
Target Audience
Primary users are performance engineers, AI/ML development teams, and DevOps/Infrastructure groups at enterprises and cloud service providers that need to optimize compiled workloads for CPUs and GPUs at scale.
Features
- Hardware‑aware LCLM that ingests CPU/GPU opcode streams, hardware counters, and register states to model performance and energy consumption at the basic‑block level
- Predictive “Shift‑Left” observability agent that flags thermal and latency spikes during pre‑test analysis
- Autonomous optimization agent that rewrites binaries and adjusts runtime configs to eliminate identified inefficiencies without human intervention
- CI/CD integration hooks that inject LOCI scans into build pipelines, providing real‑time performance budgets and failure alerts
- Runtime Resilience (RAS) module delivering dynamic power‑capping recommendations and reliability trade‑offs for production workloads
- Evidence‑based validation engine that re‑measures each change, generates quantitative reports, and tracks goal progress across releases
- RESTful API and SDKs for seamless embedding into custom tooling, monitoring dashboards, and cloud orchestration platforms