The common-pattern enrichment and tracker-mapping agents run on
reasoning models such as gpt-5-nano, which reject an explicit
temperature and fail the whole request with a 400 ("Unsupported
value: 'temperature' does not support 0.1 with this model"). The
model registry already records this capability, but nothing
consulted it before dispatch, and dated provider snapshots like
gpt-5-nano-2025-08-07 did not resolve in the registry.
Resolve dated snapshots to their undated base model in registry
Lookup, and sanitize each chat completion request in the LLM
client by omitting the sampling knobs the target model does not
accept (temperature, top_p, frequency/presence penalties, stop).
Unknown models are left untouched, so models absent from the
registry keep their current behavior.
Signed-off-by: Émile Ré <emile@probo.com>
The provider is always the prefix before "/" in the model ID,
so storing it as a separate field is redundant. Replace the
field with a Provider() method.
Signed-off-by: Aurélien Sibiril <81782+aureliensibiril@users.noreply.github.com>
Replaces []ModelDefinition with map[string]ModelDefinition so
model ID uniqueness is enforced by the data structure itself
and lookups do not require a linear scan.
Signed-off-by: Aurélien Sibiril <81782+aureliensibiril@users.noreply.github.com>
ModelDefinition, SupportedParameters, and Registry types with
multi-key lookup supporting canonical, bare, and normalized
model IDs. NewRegistry constructor accepts model definitions
for testability; DefaultRegistry caches the generated data.
Signed-off-by: Aurélien Sibiril <81782+aureliensibiril@users.noreply.github.com>
