Configuration reference¶
Every setting Logic-Loom reads, in the order the code declares them. Set them in
~/.config/logic-loom/config.toml (or the file named by LOGIC_LOOM_CONFIG), or
override one with its LOGIC_LOOM_* environment variable. Precedence and examples are
in Configuration.
| Key | Type | Default | Env var | Description |
|---|---|---|---|---|
logic_loom_root |
Path \| None |
unset | LOGIC_LOOM_LOGIC_LOOM_ROOT |
Optional Logic-Loom source checkout (display only). None = unset. |
akms_repo |
Path \| None |
unset | LOGIC_LOOM_AKMS_REPO |
Optional, development-only AKMS source checkout. None = unset; nothing requires it, in packaged builds or otherwise. When present it adds the legacy batch plan, Packages/Nodes_Vault, local graph nodes and the /api/nlm developer scripts. |
akms_workspace |
Path |
computed | LOGIC_LOOM_AKMS_WORKSPACE |
Logic-Loom-owned AKMS workspace root. Holds runtime state (prompts, workflows, batches, runs, generated nodes, NLM inputs/outputs). Bootstrapped on first sidecar startup; safe in packaged builds. Default is under Application Support / XDG data (not Documents) so macOS TCC does not block the sidecar. |
akms_global_vault |
Path \| None |
unset | LOGIC_LOOM_AKMS_GLOBAL_VAULT |
Source vault for AKMS graph compilation. None = resolve via <akms_repo>/Packages/Nodes_Vault when that optional checkout is present, then AKMS_GLOBAL_VAULT, then the AKMS default. |
learn_domain_packs_path |
Path \| None |
unset | LOGIC_LOOM_LEARN_DOMAIN_PACKS_PATH |
Root directory scanned (recursively) for AKMS Learn domain-pack descriptor YAMLs (files named 'domain_pack.yaml'). Read as DATA via akms_learn's descriptor loader to drive the Learn tab's domain-pack status UI. None = resolve via learn_domain_packs_search_root() under akms_repo. An empty or missing directory yields domain_packs: []. |
zsum_repo |
Path \| None |
unset | LOGIC_LOOM_ZSUM_REPO |
Optional, development-only zotero-summarizer checkout. None = unset; zsum runs from the installed package. |
vault_path |
Path \| None |
unset | LOGIC_LOOM_VAULT_PATH |
Published Obsidian vault root. None = unset → first-run UX. |
cache_path |
Path |
computed | LOGIC_LOOM_CACHE_PATH |
LogicLoom-owned cache (run history, port file, etc.). |
nlm_mcp_url |
str \| None |
unset | LOGIC_LOOM_NLM_MCP_URL |
NotebookLM MCP server URL for document-grounded node generation (e.g. http://localhost:3000/sse). |
nodes_folders |
list[Path] |
computed | LOGIC_LOOM_NODES_FOLDERS |
Staging directories scanned for AKMS node .md files in the Nodes Review tab. |
nodes_editor |
str \| None |
unset | LOGIC_LOOM_NODES_EDITOR |
Editor command for opening nodes (e.g. 'code'). None = system default (macOS: open). The Settings page accepts known editors with plain flags only; set any other program here in config.toml. |
port |
int \| None |
unset | LOGIC_LOOM_PORT |
Fixed port, or None for --auto-port. |
search_backend |
Literal['ripgrep', 'qmd'] |
"ripgrep" |
LOGIC_LOOM_SEARCH_BACKEND |
Default backend for /api/vault/search. |
vault_qmd_collection |
str \| None |
unset | LOGIC_LOOM_VAULT_QMD_COLLECTION |
QMD collection name registered for the vault. Required when search_backend=qmd. |
zsum_config_path |
Path \| None |
unset | LOGIC_LOOM_ZSUM_CONFIG_PATH |
Absolute path to zsum.config.yaml. None → defaults to <zsum_repo>/zsum.config.yaml at use time. |
zsum_python_cmd |
str |
computed | LOGIC_LOOM_ZSUM_PYTHON_CMD |
Command used to invoke the zsum pipeline. Defaults to running the bundled launcher through the sidecar's own Python interpreter, which works in both dev and packaged builds. Override (e.g. 'uv run zsum') only if you want to point at a different zsum install. |
zsum_cost_per_mtoken_in |
float |
3.0 |
LOGIC_LOOM_ZSUM_COST_PER_MTOKEN_IN |
USD cost per million input tokens (for cost estimation only). |
zsum_cost_per_mtoken_out |
float |
15.0 |
LOGIC_LOOM_ZSUM_COST_PER_MTOKEN_OUT |
USD cost per million output tokens (for cost estimation only). |
zsum_tokens_per_card_in |
int |
4000 |
LOGIC_LOOM_ZSUM_TOKENS_PER_CARD_IN |
Estimated input tokens per card. |
zsum_tokens_per_card_out |
int |
500 |
LOGIC_LOOM_ZSUM_TOKENS_PER_CARD_OUT |
Estimated output tokens per card. |
local_models_custom |
list[str] |
computed | LOGIC_LOOM_LOCAL_MODELS_CUSTOM |
User-added local model names shown in Home → Models. |
local_model_last_installed |
str \| None |
unset | LOGIC_LOOM_LOCAL_MODEL_LAST_INSTALLED |
Most recent local model successfully installed/configured from LogicLoom. |
anthropic_api_key |
str \| None |
unset | LOGIC_LOOM_ANTHROPIC_API_KEY |
Anthropic API key (Claude). Enables the API route for Learn LLM expansion. |
openai_api_key |
str \| None |
unset | LOGIC_LOOM_OPENAI_API_KEY |
OpenAI API key (Codex/GPT). Enables the API route for Learn LLM expansion. |
google_api_key |
str \| None |
unset | LOGIC_LOOM_GOOGLE_API_KEY |
Google API key (Gemini). Enables the API route for Learn LLM expansion. |
llm_provider |
str \| None |
unset | LOGIC_LOOM_LLM_PROVIDER |
Global host LLM provider used by every consumer (Learn/zsum/node-gen) unless a per-consumer override is set. Same vocabulary as the legacy learn_llm_provider: 'akms' (API/LiteLLM), 'claude-cli', 'codex-cli', 'nlm', or a custom label. None/'auto' = auto-resolve (nlm or API key only; CLI agents require an explicit choice). When unset, the legacy learn_llm_provider value is read as the back-compat alias. |
learn_llm_provider |
str \| None |
unset | LOGIC_LOOM_LEARN_LLM_PROVIDER |
Legacy/back-compat alias for the global llm_provider. Kept so existing config.toml files keep working untouched: when llm_provider is unset, learn_llm_provider is read as the effective global provider (effective = llm_provider or learn_llm_provider). Same vocabulary: 'akms' (API/LiteLLM), 'claude-cli', 'codex-cli', 'nlm', or a custom label. None/'auto' = auto-resolve. |
zsum_llm_provider |
str \| None |
unset | LOGIC_LOOM_ZSUM_LLM_PROVIDER |
Per-consumer override for the zsum (Zotero Summarizer) host LLM provider. None = fall back to the global llm_provider (then the learn_llm_provider alias). Same provider vocabulary as llm_provider. |
nodegen_llm_provider |
str \| None |
unset | LOGIC_LOOM_NODEGEN_LLM_PROVIDER |
Per-consumer override for the AKMS node-generation host LLM provider. None = fall back to the global llm_provider (then the learn_llm_provider alias). Same provider vocabulary as llm_provider. |
learn_llm_custom_clis |
dict[str, str] |
computed | LOGIC_LOOM_LEARN_LLM_CUSTOM_CLIS |
User-defined CLI providers: label → command template. '{prompt}' is argv-substituted (no shell); absent → prompt piped to stdin. Empty command = not registered. |
openalex_user_email |
str \| None |
unset | LOGIC_LOOM_OPENALEX_USER_EMAIL |
OpenAlex polite-pool email used as the User-Agent contact. Required for the OpenAlex client; optional only when the Assistant Discovery feature is disabled. |
openalex_api_key |
str \| None |
unset | LOGIC_LOOM_OPENALEX_API_KEY |
Optional OpenAlex API key (premium tier). |
semantic_scholar_api_key |
str \| None |
unset | LOGIC_LOOM_SEMANTIC_SCHOLAR_API_KEY |
Optional Semantic Scholar API key. The Assistant client uses the public endpoint when this is unset, with stricter rate limits. |
assistant_workspace |
Path \| None |
unset | LOGIC_LOOM_ASSISTANT_WORKSPACE |
Assistant data tree root. None = resolve via assistant_workspace_path() — either <akms_workspace>/../Assistant-Workspace, or <akms_workspace>/assistant/ when co-located with the AKMS workspace. |
assistant_workspace_colocate |
bool |
False |
LOGIC_LOOM_ASSISTANT_WORKSPACE_COLOCATE |
When True and assistant_workspace is unset, the resolver returns <akms_workspace>/assistant/ instead of the sibling Assistant-Workspace directory. Lets users keep both trees under a single root. |
failure_memory_mode |
Literal['disabled', 'read-only', 'operator'] |
"read-only" |
LOGIC_LOOM_FAILURE_MEMORY_MODE |
Permission mode for the Failure Memory workspace. 'read-only' (default) exposes inspection/preview/diagnostics only and cannot mutate any configured repository. 'operator' additionally unlocks record/refresh/compile and must be set deliberately in config.toml or LOGIC_LOOM_FAILURE_MEMORY_MODE. 'disabled' turns the surface off entirely. |
failure_memory_projects |
list[FailureMemoryProject] |
computed | LOGIC_LOOM_FAILURE_MEMORY_PROJECTS |
EXPLICITLY configured failure-memory projects, as [[failure_memory_projects]] array-of-tables entries with id + config (+ optional root). Logic-Loom never discovers .failure-memory directories by walking the filesystem — an unlisted project is invisible to the app, by design. |
records_stores |
list[RecordsStore] |
computed | LOGIC_LOOM_RECORDS_STORES |
EXPLICITLY configured akms-records stores, as [[records_stores]] array-of-tables entries with id + scope (+ repo for a project scope, optional home for agent/shared). Logic-Loom only reads them and never discovers stores by scanning. |
seed_starter_batches |
bool |
True |
LOGIC_LOOM_SEED_STARTER_BATCHES |
Seed the AKMS batch store with shipped starter batches on first sidecar startup. Seeding runs only when the store is empty; existing batches (including any whose id collides with a seed id) are never overwritten. Set False to skip seeding entirely — useful for clean-room testing or users who want a blank workspace. |