Module domain

Expand description

§domain

Core domain types.

These types represent the pure domain model, independent of any infrastructure concerns (database, filesystem, etc.).

§Structure

agent - Agent loop types (AgentConfig, AgentMessage, AgentEvent, etc.)
model - Model types (Model, NewModel)
model_naming - Shared model-naming policy (resolve_model_name, NameSource)
mcp - MCP server types (McpServer, NewMcpServer, etc.)
chat - Chat conversation and message types
gguf - GGUF metadata and capability types
capabilities - Model capability detection and inference
thinking - Thinking/reasoning tag parsing and streaming accumulation
kv_memory - Shape of a model’s KV memory from GGUF metadata: whether it keeps only part of the token history (kv_memory_is_partial), and how many layers hold a per-token cache at all (kv_cache_layer_count)
kv_estimate - Per-token KV cache size, which takes that layer count from kv_memory rather than counting every block. The dependency runs one way only: kv_memory reads metadata and knows nothing of the estimate

Modules

Module	LOC	Complexity	Coverage
`admission.rs`
`cache_budget.rs`
`capabilities.rs`
`capability_tags.rs`
`chat.rs`
`context_fit.rs`
`defect_counts.rs`
`defects.rs`
`dialect.rs`
`generation_config.rs`
`gguf.rs`
`inference.rs`
`inference_profile.rs`
`inference_profile_tests.rs`
`inference_tests.rs`
`kv_estimate.rs`
`kv_memory.rs`
`kv_memory_tests.rs`
`launch_narration.rs`
`loop_guard_log.rs`
`model.rs`
`model_naming.rs`
`model_sampling.rs`
`query.rs`
`reasoning_effort.rs`
`reasoning_effort_tests.rs`
`recommendation.rs`
`residency.rs`
`runtime_capabilities.rs`
`sampling_discards.rs`
`sampling_discards_non_finite_tests.rs`
`sampling_discards_tests.rs`
`sampling_provenance.rs`
`sampling_provenance_tests.rs`
`server_config.rs`
`slot_eviction.rs`
`template_caps.rs`
`template_caps_tests.rs`
`agent/`
`benchmark/`
`mcp/`

Re-exports§

pub use benchmark::BenchmarkEvent;
pub use benchmark::BenchmarkModelResult;
pub use benchmark::BenchmarkRun;
pub use benchmark::BenchmarkRunStatus;
pub use benchmark::BenchmarkRunType;
pub use benchmark::CandidateSource;
pub use benchmark::CompareConfig;
pub use benchmark::GeneratedOutput;
pub use benchmark::ModelBenchmarkSummary;
pub use benchmark::ModelCompareResult;
pub use benchmark::ModelPerfResult;
pub use benchmark::PerfConfig;
pub use benchmark::ScoreWeights;
pub use benchmark::SweepSpec;
pub use benchmark::TaskCategory;
pub use benchmark::TaskSuite;
pub use benchmark::TuneCandidateResult;
pub use benchmark::TuneConfig;
pub use benchmark::TuneTask;
pub use benchmark::TuneTaskResult;
pub use capability_tags::is_reasoning;
pub use inference_profile::InferenceProfile;
pub use inference_profile::builtin_templates;
pub use recommendation::BUDGET_UTILISATION;
pub use recommendation::BudgetSource;
pub use recommendation::Recommendation;
pub use recommendation::recommend;
pub use slot_eviction::SlotFileMeta;
pub use slot_eviction::compute_auto_disk_budget_bytes;
pub use slot_eviction::select_evictions;
pub use mcp::McpEnvEntry;
pub use mcp::McpLifecycle;
pub use mcp::McpServer;
pub use mcp::McpServerConfig;
pub use mcp::McpServerStatus;
pub use mcp::McpServerType;
pub use mcp::McpTool;
pub use mcp::McpToolResult;
pub use mcp::NewMcpServer;
pub use mcp::ToolIndex;
pub use chat::Conversation;
pub use chat::ConversationUpdate;
pub use chat::Message;
pub use chat::MessageRole;
pub use chat::NewConversation;
pub use chat::NewMessage;
pub use gguf::CapabilityFlags;
pub use gguf::GgufCapabilities;
pub use gguf::GgufMetadata;
pub use gguf::GgufValue;
pub use gguf::RawMetadata;
pub use dialect::BodyCodec;
pub use dialect::DialectSpec;
pub use dialect::EmissionProfile;
pub use agent::AGENT_EVENT_CHANNEL_CAPACITY;
pub use agent::AgentConfig;
pub use agent::AgentEvent;
pub use agent::AgentMessage;
pub use agent::AssistantContent;
pub use agent::DEFAULT_MAX_ITERATIONS;
pub use agent::DEFAULT_MAX_STAGNATION_STEPS;
pub use agent::LlmStreamEvent;
pub use agent::LoopDetector;
pub use agent::MAX_ITERATIONS_CEILING;
pub use agent::MAX_PARALLEL_TOOLS_CEILING;
pub use agent::MAX_TOOL_TIMEOUT_MS_CEILING;
pub use agent::StagnationDetector;
pub use agent::ToolCall;
pub use agent::ToolDefinition;
pub use agent::ToolResult;
pub use capabilities::capabilities_from_architecture;
pub use capabilities::infer_from_chat_template;
pub use capabilities::ChatMessage;
pub use capabilities::MessageContent;
pub use capabilities::ModelCapabilities;
pub use capabilities::transform_messages_for_capabilities;

Modules§

admission 🔒: What the admission queue and the VRAM resident set look like right now.
agent: agent
benchmark: benchmark
cache_budget 🔒: Auto-sizing math for llama-server’s host-RAM prompt cache (--cache-ram).
capabilities: Model capability detection, inference, and request transformation.
capability_tags: The capability tags gglib stores on a model, and the predicates that read them.
chat: Chat domain types.
context_fit 🔒: How large a context this machine can actually serve.
defect_counts: The per-model defect counts — the data half of super::defects.
defects: Per-model defect counters — the Tier C signals the closed loop steers by.
dialect: Tool-call dialect specification.
generation_config 🔒: The sampling recipe a model author publishes in generation_config.json.
gguf: GGUF domain types.
inference 🔒: Inference configuration types.
inference_profile: Named, cross-model sampling profiles.
kv_estimate 🔒: KV-cache size estimation from GGUF metadata.
kv_memory 🔒: The shape of a model’s KV memory, read from GGUF metadata: whether it retains the full token history, and how many layers hold a cache at all.
launch_narration 🔒: What the runtime decided for one llama-server launch, and why.
loop_guard_log: The loop guard’s log, which outlives the process.
mcp: mcp
model 🔒: Model domain types.
model_naming 🔒: Shared model-naming policy.
model_sampling 🔒: The sampler defaults a model carries in its own GGUF metadata.
query 🔒: Model list query, filter, and sort types.
reasoning_effort 🔒: The reasoning_effort level — how hard the model is asked to think.
recommendation: Picking a first model that actually fits the machine it will run on.
residency 🔒: Whether a second model may stay resident in VRAM alongside the first.
runtime_capabilities 🔒: What the engine underneath gglib can do — the mirror of GgufCapabilities.
sampling_discards 🔒: Which parameters a caller named and the ladder then passed over.
sampling_provenance 🔒: Which layer supplied each resolved sampling parameter.
server_config 🔒: Server-level default configuration for models.
slot_eviction: Byte-budget selection for on-disk KV slot cache eviction.
template_caps 🔒: llama-server’s per-template capability self-report, and the tri-state gglib holds it in.

Structs§

AdmissionSnapshot: Everything the admission queue and resident set look like right now.
FieldSources: Per-field provenance for one resolved InferenceConfig.
FitInputs: What fit_context worked from, for a person reading a launch log.
InferenceConfig: Inference parameters for LLM sampling.
KvElemsPerToken: Per-token K and V element counts, type-agnostic.
LaunchDecision: One resolved launch decision, paired with the reason it was chosen.
LaunchNarration: Everything the runtime decided for one llama-server launch.
Model: A model that exists in the system with a database ID.
ModelFile: Represents a single file (shard) belonging to a model.
ModelFilterOptions: Filter options for the model library UI.
ModelListQuery: Complete filter + sort specification for the model list.
ModelSamplingContext: Everything about the target model that changes how sampling resolves, independent of any specific request.
ModelSamplingDefaults: The sampler defaults one model declares.
NewModel: A model to be inserted into the system (no ID yet).
NewModelFile: A model file entry to be inserted into the system (no ID yet).
QueuedModelSnapshot: Requests waiting for one model that is not currently resident.
RangeValues: A range of numeric values with min and max.
ResidentSlotSnapshot: One model resident in VRAM.
RuntimeCapabilities: What the llama-server binary underneath gglib is, and what it can do.
RuntimeFlags: Capabilities of the llama-server binary gglib is running against.
SecondarySlotStatus: Why the second VRAM slot is or is not in use.
ServerConfig: Server-level defaults for a specific model.
SlotFootprint: What one resident model is expected to occupy in VRAM.
TemplateCaps: The nine bools of chat_template_caps, as GET /props reports them.

Enums§

CacheRamHealth: How healthy a resolved --cache-ram budget is, for user-facing display.
DefaultsOrigin: Whether a model’s stored inference_defaults were set by the user or written automatically at import time.
FieldIssue: What reading one client-supplied sampling field did, when it was not simply “read it”.
ModelSamplingDefault: What one model’s GGUF says about one sampler field.
ModelSortBy: The field to sort the model list by.
NameSource: Which naming signals are available for a model being added.
ParamSource: Which rung of a sampling ladder supplied one resolved parameter.
ReasoningEffort: A reasoning-effort level gglib is willing to put on the wire.
SamplingLayer: The five rungs of the ladder resolve_with_profile builds, in priority order.
SamplingOverride: What gglib is doing with one field’s published recommendation.
SecondarySlotDecision: Whether a candidate may take the second resident slot, and why not when it may not.
SortOrder: Direction for sorting.
Support: One capability’s answer, with unknown kept distinct from no.
TemplateCapsState: What gglib currently holds about a model’s template caps.

Constants§

CACHE_RAM_UNKNOWN_KV_ALLOWANCE_BYTES: KV allowance assumed when the model’s metadata doesn’t permit an estimate. Deliberately generous: over-reserving shrinks the cache (safe), whereas under-reserving risks memory pressure.
MODEL_SAMPLING_KEYS: gglib’s wire name for a sampler field, paired with the GGUF key that can move it.
RESIDENCY_UTILISATION: Fraction of free VRAM a co-resident candidate is allowed to claim.

Functions§

apply_query: Apply a ModelListQuery to a model list, returning a filtered and sorted copy.
classify_cache_ram: Classify a resolved --cache-ram budget.
compute_auto_cache_ram_mb: Compute the auto --cache-ram budget, in MiB.
decide_secondary_slot: Decide whether candidate may stay resident alongside what is already loaded.
declared_name: The general.name declared in the GGUF header, or None if absent or blank.
discarded_from_rung: The parameters named set that did not survive to the resolution.
estimate_kv_bytes_for_context: Estimate total KV cache bytes for a given context size.
estimate_kv_elems_per_token: Estimate K and V element counts consumed per token of context.
fit_context: The largest context weights_bytes can serve inside budget_bytes.
fit_context_explained: The same calculation, reporting what it worked from.
format_gib: Format a byte count as GiB with one decimal, e.g. 17.2 GiB.
format_mib_as_gib: Format a MiB count as GiB with one decimal, for the RAM cache budget.
generation_config_candidates: Where to look for a generation_config.json, best candidate first.
is_system_tag: Returns true when tag is a system tag that callers must not remove through the standard tag-mutation API.
kv_bytes_per_token: Convert per-token K/V element counts to bytes at the given cache types.
kv_cache_layer_count: How many of the model’s layers hold a per-token KV cache.
kv_memory_is_partial: Whether the model’s KV memory retains only part of the token history (sliding-window, hybrid, or recurrent attention).
parse_generation_config: Read a model author’s generation_config.json.
reasoning_effort_support: Whether a model’s template reads reasoning_effort, from its recorded caps.
repo_short_name: Strip a HuggingFace repo id down to its final path segment.
resolve_model_name: Resolve the models.name for a model being added.
strip_gguf_suffix: Strip a trailing -GGUF marker, case-insensitively.

Module domain

Module domain Copy item path

§domain

§Structure

Modules

Re-exports§

Modules§

Structs§

Enums§

Constants§

Functions§

Module domain