Start with one reply.
Build a complete desktop agent.

Meet pi-ai, providers, models, and context through one minimal model call.

Understand AssistantMessageEventStream and streaming events.

Turn model internals into states a desktop application can display.

Understand providers, the model catalog, authentication, and cross-provider compatibility.

Learn ToolCall and ToolResult by building a file-information tool.

Connect model decisions, tool execution, and returned results into a loop.

Map agent events to clear, observable interface states.

Handle concurrent execution, aborts, errors, and stopping conditions.

Define permission, approval, trust, and sandbox boundaries.

Separate agent messages, LLM messages, UI messages, and context.

Understand sessions, entries, repositories, and persistence.

Learn forks, branches, and retrying from a historical node.

Understand overflow, cut points, summaries, and compaction records.

Use steering, follow-ups, and queues to control active work.

Understand dynamic prompts, project context, and resource merging.

Distinguish Skills, tools, system prompts, and Extensions.

Build a runnable error-log analysis Skill.

Trace directory scans, frontmatter parsing, and conflict diagnostics.

Understand explicit invocation, automatic matching, and progressive loading.

Create Skills that are discoverable, executable, verifiable, and safe.

Cover tests, package distribution, conflicts, and security governance.

Use tool approval to understand event-driven extension.

Trace discovery, module loading, factory registration, and runtime binding.

Understand registration, execution, and AgentSession bindings.

Follow observation, modification, and blocking events through one request.

Understand tool definitions, wrappers, approval, and auditing.

Choose extension points for prompts, input, models, compaction, and sessions.

Extend models, authentication, Skills, resources, sandboxes, and services.

Handle errors, cancellation, conflicts, reloads, concurrency, and compatibility.

Compare pi-ai, agent-core, the harness, SDK, and RPC.

Organize an application with AgentSession, services, and runtime components.

Design the event protocol between a desktop process and a Pi child process.

Map conversation, streaming, tools, permissions, and sessions into UI state.

Connect chat, tools, memory, Skills, and Extensions.

Follow input through model output, tool execution, and session persistence.

Complete security, recovery, evaluation, diagnostics, packaging, and upgrade checks.

Learn the new capabilities, TypeBox breaking changes, and a verifiable migration path.