Agent harnesses
Pi Agent Harness: A Reading Path
Ten readings on Pi's tool protocol, loop, sessions, compaction, extensions, and product choices, with a source-grounded study sequence.
Technical writing
Notes
Notes on agent harnesses, inference serving, concurrency, queues, dependency ordering, and small stateful systems. The collection includes reading paths, worked implementations, and compact references.
Reading Paths
Curated source material, arranged in the order that makes the system easiest to understand.
Source Studies
Implementation traces, design boundaries, and exercises grounded in a pinned source snapshot.
Agent harnesses
Pi Agent Harness: Source Study Notes
Pi's event stream, loop, tools, session tree, compaction, extensions, tmux workflow, and security boundary, with five integrated exercises.
Guides
Inference serving, concurrency, queues, shutdown, worker pools, and asyncio.
Inference systems
LLM Inference Serving: Essential Notes
Prefill, decode, KV-cache memory, prefix-aware routing, eviction, memory tiers, continuous batching, PagedAttention, and the scheduling tradeoffs connecting them.
First principles
Python Concurrency From First Principles
Threads, shared state, conditions, semaphores, blocking queues, shutdown, worker pools, testing, and asyncio, built from concrete questions about who can proceed and why.
Worked Labs
One worker pool, built and traced from submission through shutdown.
Execution trace
A Small Python Worker Pool, Explained by Execution
A bounded worker pool traced through queue state, unfinished-work accounting, sentinel shutdown, failures, backpressure, tests, and futures.
References
Python APIs, invariants, operation costs, and implementation patterns.
Pocket reference
Python Systems Coding Reference
A compact map of core data structures, sorted lookup, dependency ordering, concurrency primitives, blocking queues, thread pools, asyncio, and common failure checks.