Expand description
Catalog LLM download → GGUF shard map → WAL → lifecycle (in-process).
Structs§
- Active
Model Record - Install
Manifest - Model
Install Result - Model
Status - Vault
Duplicate Group - Exact duplicate GGUF files, grouped by content rather than filename.
- Vault
Gguf Entry - Discovered weight container in a vault directory (CLI / lifecycle tooling).
Enums§
Constants§
- MEMORY_
FLOOR_ MB - UI RAM ceiling for LLM arena pressure (matches 0.0.12 Flutter plan).
Functions§
- activate_
model - Activates a model.
- activate_
model_ for_ id - Activates a model by ID.
- activate_
vault_ gguf - Map and activate a vault weight file (
.p64preferred, or.gguf) without catalog install. - audit_
vault_ duplicates - Hash only same-sized GGUF candidates and report byte-identical duplicates.
- finalize_
llm_ install - Finalizes an LLM install.
- finalize_
local_ gguf - Register and activate a user-selected
.gguffile at an arbitrary path. - get_
kv_ cache_ used_ mb - Gets kv-cache-used-mb.
- get_
last_ decode_ tok_ s - Last measured tok/s from a completed turn, or
Noneif never measured this process. - get_
last_ decode_ tok_ s_ at_ unix - Gets last-decode-tok-s-at-unix.
- get_
llm_ memory_ bytes - Gets llm-memory-bytes.
- get_
model_ lifecycle_ state - Gets model-lifecycle-state.
- get_
model_ status - Gets model-status.
- get_
thermal_ state_ label - Gets thermal-state-label.
- install_
catalog_ llm - Installs a catalog LLM.
- lifecycle_
label lifecyclelabel.- load_
install_ manifest - Loads install-manifest.
- models_
dir - Models directory.
- record_
kv_ cache_ used_ mb - Records KV-cache usage.
- record_
last_ decode_ tok_ s - Record measured decode/chat throughput from a completed generation turn.
- record_
llm_ memory_ bytes - Records LLM memory bytes.
- record_
llm_ memory_ sample - Records an LLM memory sample.
- resolve_
vault_ model - Resolve a model reference (filename, stem, or path) inside
vault_dir. - scan_
vault_ gguf - Recursively scan
vault_dirfor.p64and.ggufweight containers. - task_
orchestrator - Shared orchestrator for chat inference and model lifecycle transitions.
- unload_
active_ model - Unloads the active model.
- wait_
for_ eviction_ scrub - Block until eviction scrub completes or
timeoutelapses.