DeepCode Guides

Task-oriented, teaching-style documentation. Every command, flag, and output shown here is taken from the current source tree, not from memory — each page names the module that makes it true.

Guide	You will learn
Getting started	Install, connect a model, trust a folder, run your first session — and read what the agent shows you
The terminal UI	Every slash command, pickers, @file and $skill references, transcript detail, interrupting, approvals
Sessions	Resume, work across two windows, compact a long conversation, what lives on disk
Models and providers	Connections, declared models, switching mid-conversation, reasoning effort
Skills and memory	Reusable playbooks, project instructions, durable notes, plugins
Goals, automation, and headless runs	Durable goals, the verify-loop, deepcode exec --json, schedules, MCP serving

Two other surfaces share everything these guides describe:

• Desktop — a Tauri app over the same runtime. A Desktop thread and a CLI session are the same record with the same id; start in one, continue in the other.
• Headless — deepcode exec runs one task and exits; --json streams structured events for scripts and CI.

Reference documents (architecture, contracts, deep dives) remain in docs/. The Paper2Code research pipeline keeps its own documentation: see the README’s Paper2Code section.