Agnitra Platform
Agnitra gives your team an opinionated workflow for profiling, optimizing, and shipping large language models with built-in telemetry and billing. The SDK and CLI handle model compilation, surface runtime hotspots, and emit marketplace-ready usage events so you can move from prototype to production without standing up separate infrastructure.Platform Pillars
- Optimization agents adapt TorchScript/ONNX graphs with LLM + RL tuned kernels so you keep model accuracy while boosting throughput.
- Telemetry-first workflows capture latency, GPU hours, and savings in structured JSON so finance and infra teams share the same source of truth.
- Usage-based monetization maps optimization runs to Stripe, AWS Marketplace, and internal ledgers with auditable metadata and license enforcement.
Start Building Fast
- Install the SDK from PyPI and ship your first optimized artifact via the Quickstart.
- Explore repeatable automations and code samples in the SDK & CLI Guide.
- Understand what telemetry and usage events look like in practice with the Telemetry Playbook.
Architecture Overview
The MVP couples a telemetry collector, FX graph extractor, AI optimizer, kernel generator, and runtime patcher behind a unified CLI and control plane. Read the Architecture Deep Dive for module responsibilities, data flows, and performance targets drawn from the PRD.Key Workflows
- Monitor production uplift and route usage events using the Telemetry Playbook.
- Connect finance tooling and marketplaces through the Marketplace & Billing guide.
- Bring the control plane into your infrastructure with the
agnitra-apireference.
Reference Materials
- Configure environments with the Runtime Configuration reference.
- Integrate the OpenAI Responses API using the Responses API contract.
- Mirror telemetry exports and SDK helpers with code snippets across the repository (see
agnitra/core,agnitra/api, andcli/).
What’s Next?
Runpip install agnitra, export AGNITRA_API_KEY, and launch agnitra optimize against the TinyLlama fixture to validate your workstation. From there, follow the linked guides to embed Agnitra into pipelines, dashboards, and billing loops.