Modules§
Structs§
- Lifecycle
Options - Plan
Options - Quant
Plan - Speculation
Plan - How a launch will speculate, decided before any side effect.
Functions§
- auto_
backend_ order - Backends compatible with the model, in auto-selection order; MLX leads only on Apple Silicon. Attach-only LM Studio is never auto-selected (spec Q1).
- check_
attach_ override - Attach-intent commands follow the active backend; a conflicting override must fail, not be ignored.
- check_
backend - drafter_
reserve hardwarewith the drafter’s catalog memory and download ceiling set aside.- forget_
attached_ with_ config down --forget: clears a stale attached pointer without probing or signalling any process.- installed_
inventory - kv_
slot_ directory - Where llama.cpp servers rigspark starts keep saved KV slots:
kv-slots/beside the runtime state file, created private to this user. - kv_
slot_ directory_ for - The KV slot directory a server of
backendstarts with: llama.cpp only. - kv_
slot_ with_ config - Saves, restores or erases a KV slot on the llama.cpp server this config’s state records. Lets a caller free a slot during a long pause (a tool call) and bring it back after.
- plan_
quantization - Picks the quantization to pull and enforces fit, bypass and disk preflight before any side effect.
- plan_
speculation - Resolves
auto(and the llama.cpp default when no flag was given) to a concrete mode. - preferred_
backend - Flag, then non-blank
RIGSPARK_BACKEND, then the user config default. - run_
native - run_
native_ observed - run_
native_ with_ config - run_
native_ with_ config_ observed - slot_
sizing_ model - The model as sized for concurrent slots: Ollama allocates a full context per slot, llama.cpp shares one unified KV pool.
- validate_
backend_ platform