Agents That Own Their Inference

AI-Native Workshop is a repository-based workshop for practicing AI-native software development workflows. Its curriculum covers voice coding, agent sessions, loops and measurable goals, Git worktrees, verification gates and hooks, adversarial review, and scheduled tasks. The repository includes setup skills, exercises, slides, a glossary, a facilitator guide, and an optional local `aie-coach` MCP server with seven tools for check-ins and setup scoring. Its setup workflow runs inside Claude Code and can install and verify Bun, the Codex CLI, and Git; the workshop is organized in four blocks covering voice, loops and goals, gates, and schedules.

The Mastering AI Observability Workshop repository is a hands-on workshop resource for building a trace-backed support-agent improvement workflow with Braintrust. It provides a local Python workshop app, setup instructions, a support agent using tool calls, seeded evaluation cases, production-like trace generation, Braintrust datasets and evaluations, hosted scorers, Topics facets, sample-trace imports, and optional draft evaluation cases derived from traces. The workflow connects tracing and online analysis to offline evaluations, custom scoring, failure investigation, regression cases, and reviewable code changes. The repository includes a public WORKSHOP.md guide, environment setup through uv and Make commands, local database seeding, a web interface, and configuration for Braintrust Gateway model calls.

FriendliAI is an inference cloud and API for running frontier open-weight and custom AI models in production, with support for long-context inference, real-time streaming, tool calling, and agent workloads. Its stack combines custom GPU kernels, continuous batching, parallel inference, prefix caching with cache-aware routing, and a GPU-, CPU-, and NVMe-based KV-cache hierarchy for repetitive or long agent inputs. It also uses hybrid speculative decoding that adapts to live traffic, alongside multi-cloud scaling and autoscaling. The service provides model APIs and one-click deployment for models from Hugging Face, as well as deployment support for proprietary or fine-tuned models. FriendliAI advertises 99.99% uptime SLAs and reports performance claims including up to 7x faster output token speed and lower inference costs.

Agent Context Workshop is a hands-on workshop resource for building and evaluating context layers for coding agents. It uses selected crates from the Qdrant Rust codebase to extract syntax with tree-sitter, resolve symbols and relationships with rustdoc JSON, store the resulting code knowledge graph in Neo4j, and expose graph tools such as search, blast-radius analysis, and implementor lookup to an agent. The workshop compares graph-based agents with grep-style agents using gold-verified questions, deterministic scoring, cached evaluation results, tool-adoption measurements, and trace inspection. Additional notebooks cover static Python resolution with Griffe and an optional Fenic-and-LLM semantic layer. The GitHub repository maintains the workshop through ordered Jupyter notebooks, typed extraction and graph code, evaluation harnesses, tests, committed Qdrant seed data, and a committed evaluation cache; it is licensed under Apache-2.0.

vLLM is an open-source library and inference-serving engine for deploying large language models on GPUs and other hardware accelerators. Originally developed in UC Berkeley's Sky Computing Lab, it is maintained by a broad community and provides an OpenAI-compatible API server, as well as Anthropic Messages API and gRPC support. Its serving stack manages attention key-value memory with PagedAttention, continuous batching, chunked prefill, prefix caching, streaming generation, and disaggregated prefill, decode, and encode. It also supports speculative decoding, quantization, optimized attention and GEMM/MoE kernels, automatic kernel generation and graph-level transformations, structured outputs, tool calling, multiple decoding algorithms, and distributed inference through tensor, pipeline, data, expert, and context parallelism. The project integrates with Hugging Face models and supports decoder-only, mixture-of-experts, hybrid state-space, multimodal, embedding, retrieval, reward, and classification models. It can run across NVIDIA, AMD, Intel, and other supported accelerators, as well as x86, ARM, and PowerPC CPUs, and can be installed with uv or pip or built from source.