Skip to content

LLM providers

Logic-Loom has one LLM setting that drives all three things that need a model: AKMS Learn, the zsum pipeline and AKMS node generation. Set it once under AKMS ▸ Settings, or as llm_provider in config.toml.

llm_provider What it uses
akms API keys through LiteLLM (anthropic_api_key, openai_api_key, google_api_key)
claude-cli The Claude Code CLI on your machine
codex-cli The Codex CLI
nlm NotebookLM through the nlm CLI
a custom label A CLI you register under learn_llm_custom_clis
unset or auto Picks nlm or an API key automatically; CLI agents need an explicit choice

API keys are stored in config.toml and never sent back to the app window: a saved key shows as a row of dots in Settings. Type over it to replace it, or empty the field to remove it.

Per-feature overrides

To use a different provider for one feature, set an override. Each falls back to llm_provider when unset.

llm_provider = "claude-cli"
zsum_llm_provider = "akms"        # zsum uses an API key instead
# nodegen_llm_provider = "nlm"

Older config files that set learn_llm_provider keep working: it is read as the global provider when llm_provider is unset.

Custom CLIs

Register any command-line model as a provider. {prompt} is replaced with the prompt as one argument (no shell); without it, the prompt is piped to stdin.

[learn_llm_custom_clis]
local-llama = "llm -m llama3 {prompt}"

Local models

Home ▸ Models can install and configure a local Ollama model. From the bundled zotero-summarizer CLI, zsum setup-model --size small|medium|large|xlarge does the same in one command.

Logic-Loom never stores provider sign-ins. Each CLI uses its own login.