Glossary¶
Terms this project uses precisely. Where a word has a loose industry meaning and a specific meaning here, the specific one is what the code implements.
Adapter¶
The per-provider module that translates a WireRequest into a provider's wire format and
its responses back into events. Adapters only translate; the boundary is explained in
writing a provider adapter.
Alias¶
A tier name (small, medium, large) that resolves to a concrete model per provider
through the catalog; see targets and aliases.
Arena¶
Running one request against several targets at once and picking a winner by a declared strategy, rather than falling back only on failure; see arena.
Attempt¶
One try against one resolved target. A request may involve several, across retries and fallback; the full attempt trail is on every result.
Capability¶
Something a model can do, paired with the provenance of how that is known; see capabilities and provenance.
Cassette¶
Recorded HTTP traffic replayed in tests, so conformance can run without credentials; recording one takes a single command.
Catalog¶
The data file mapping aliases to per-provider targets, and artifact ids to pinned, hash-verified downloads; see the model catalog.
Confidentiality Tier¶
One of five numbered postures (0-4) for protecting prompt IP, each stating its own guarantee and its ceiling; see confidentiality tiers.
Conformance Suite¶
The shared test suite every adapter must pass before its matrix row is published; see the conformance suite.
Descriptor¶
Frozen, declarative data about a provider: how to build its adapter, what configuration it needs, how it spells reasoning effort, what it supports. See start from a descriptor.
Drift Check¶
The semi-automated audit comparing contract snapshots against providers' current public
documentation, following
contracts/DRIFT-CHECK.md.
Event Stream¶
The generation primitive. A generation is an ordered stream of typed events, described in the event stream; the non-streaming API drains it.
Extra¶
An optional install group (anyinfer[serve], [otel], [demo]) naming a capability
rather than a provider, so the core install stays two dependencies wide. The full table
is in installation.
Health Gate¶
Skipping a target that recently failed, for a short TTL, so one dead endpoint does not cost every request its full timeout; see health gating.
Mechanism¶
How structured output was requested: grammar,
json_schema, json_mode, or prompt. Recorded on every result.
Posture¶
How much of a machine local inference may commit:
conservative, balanced, or aggressive.
Preset¶
A named configuration of the shared openai_compat adapter for one OpenAI-compatible
service — a registry entry, not a dedicated adapter; see
presets.
Projection¶
Rewriting a schema for a provider's wire format (stripping keywords a grammar compiler cannot handle efficiently). Never changes what validation checks client-side.
Provenance¶
Where a capability value came from: default, catalog, discovered, probed, or
override, weakest to strongest; an
application's override outranks everything the library collected.
Reduction¶
Selecting and packing a caller-approved corpus down to a token budget before dispatch, with the application deciding what may be sent; see context reduction.
Repair¶
Re-prompting the same model with validation errors after a schema violation, within a bounded budget; see repair.
Route¶
An ordered list of targets plus policy: retries, health gating, and failure-class-specific chains. See routing and rate limits.
Run Manifest¶
The record of what one call actually did — targets attempted, mechanism chosen, spend, timings — available even when the call failed; see run manifests.
Sentinel Model¶
A model id meaning "the provider chooses" (Copilot's auto). Capabilities for one are
the conjunction across every candidate.
Session¶
An opaque, target-bound handle that lets capable providers reuse server-side state across calls. It never changes an answer, and reuse is reported rather than assumed; see sessions.
Sidecar¶
The OpenAI-compatible HTTP service, anyinfer serve. A wire codec
around a normal client, never a second core.
Target¶
Where a request goes: an alias, or provider:model.
TTFT¶
Time to first token, measured by the core at the first content event, identically for every provider.
Wire Request¶
A fully-resolved request handed to an adapter: concrete model, chosen mechanism, projected schema, translated reasoning effort, merged options. See the adapter contract.