Getting started

The memory and workspace layer for AI coding agents. archeus sits in front of the agent rather than replacing it, and does not proxy your conversations: it configures the agent, launches it with the model, effort, permissions and context you meant, and maintains the project's context between sessions. Claude Code is the agent it drives today.

Concretely, that is four jobs:

Memory	A semantic graph of the codebase, injected through three token-budgeted surfaces so a fresh session starts already knowing the project.
Workspace	Every folder the agent has ever opened, every session inside it — browsable, searchable, taggable, forkable, archivable, across every account.
Launch control	Model, reasoning effort, permission mode, subagents, worktree, account — chosen per project and remembered.
Cost	What a turn costs across every surface at once, and the tools to cut it.

It is pure Python standard library with zero runtime dependencies, and it uses the Claude Code authentication you already have. No API key.

The guided tour

Both surfaces ship a tour of themselves, and it is the fastest way through everything below. In the desktop app it runs as an overlay that opens each screen as it describes it — the ? at the foot of the sidebar, then Guided tour. In the terminal UI it is Getting started on the main menu. Two lengths: the first five minutes, and every function archeus has with what each one is for.

You can also walk it in the browser, with the screenshots, at claudectl.space/getting-started — same steps, same words, no install required.

Where this is going

Half of it has arrived: a session can be routed at any endpoint serving POST /v1/messages — a local Ollama, llama.cpp or vLLM server, OpenRouter, or OmniRoute — and it stays a real Claude Code session. See Model providers.

The other half has started. archeus also reads OpenAI Codex and pi — their sessions, projects, previews, models and token spend all merge into the same lists, one memory graph is delivered to all three, and any of them can be launched from the same picker. See More than one CLI.

Claude Code is the first surface, not the boundary — and it is no longer the only one. It remains much the deepest of the three, and the comparison table on that page says plainly what each of the others cannot do rather than hiding the gaps. A harness of archeus's own remains a long-term goal, not a feature.

The three surfaces

The same engine, three ways in. Pick whichever you prefer — they do the same things.

• Terminal UI — archeus

  Keyboard-first, one screen per job, instant. The default. Every action has a key and ? prints the map.

• Desktop app — archeus --gui

  The whole workspace as a local app on loopback, with the dashboard, the usage banner and the theme system. Full parity with the terminal UI.

• Claude Code plugin — /plugin install

  Three slash commands and eight skills inside a session, for recall, status and review without leaving the conversation.

The command line is the fourth way in and the one scripts and hooks use — no UI, one answer on stdout.

Where to go next

Installation	pipx, pip, a checkout, the desktop window, Windows shortcuts
Quickstart	install to first session in five minutes
Configuration	every file archeus reads and writes, and where
Projects	health checks, auto-fixes and whether the generated context still matches the repo
Sessions	browse, search, tag, fork, resume, archive, export
Project memory	the memory graph and its three injection surfaces
Architecture graph	the interactive dependency view
Usage & cost	what you spend per turn, and how to cut it
Troubleshooting	when something does not work

Reference material — the HTTP API, multiple accounts, MCP servers, agents & skills, hooks, Plan → Execute, the status line — is in the Reference section of the sidebar.