← Docs hub

Pick your mode

llmwiki synthesis backends share the same three-layer pipeline (raw/ → wiki/ → site/) but differ on who calls the LLM:

When to pick which

• You use Claude Code daily: set synthesis.backend: claude (see Agent mode).
• You want fully local models: set synthesis.backend: ollama — Tutorial 08.
• You're evaluating offline: leave dummy (default).

The old agent-delegate pending-prompt mode (--list-pending / --complete) was removed in v1.4.0. Anthropic HTTP batch / API-mode scaffolding was also removed — use claude or ollama instead.

The backends share

Everything except synthesis:

• Adapters (claude_code, codex_cli, cursor, gemini_cli, copilot_chat, obsidian, …) work identically.
• The static site, graph viewer, lint rules — all backend-agnostic.
• config.json / sessions_config.json is the same file; only synthesis.backend differs.

Read next

• Claude CLI / agent workflow
• Configuration — synthesis backend
• Upgrade guide