All pull requests

Pull requests

• feat(batch): batch inference for internal users, with per-model caps admins change live

  #1651·1a1a11a opened Oct 25, 2026Contributor·3 tasks done·

  ·
  9

• fix(audit): retain actors and distinguish typed targets

  #1650·realtmxi opened Oct 25, 2026Collaborator·3 tasks done·

  ·

• fix(routing): size the streaming engine wait on the request's own un-cached prefill

  #1637·B-A-M-N opened Oct 23, 2026Contributor·

  ·
  2

• fix(routing): let verified cache locality refine the weighted draw

  #1636·B-A-M-N opened Oct 23, 2026Contributor·

  ·
  2

• fix(registry): make include_usage_in_stream a per-route streaming capability

  #1635·B-A-M-N opened Oct 23, 2026Contributor·

  ·
  1

• refactor(routing): split backend implementations into focused modules

  #1626·B-A-M-N opened Oct 22, 2026Contributor·3 tasks done·

  ·

• refactor(serving): move recent-request query and presentation out of user_routes

  #1625·B-A-M-N opened Oct 22, 2026Contributor·3 tasks done·

  ·
  1

• fix(serving): translate tool_choice none and disable_parallel_tool_use (issue #876 P3)

  #1613·B-A-M-N opened Oct 21, 2026Contributor·

  ·
  3

• fix: expose served-identity X-FI-* response headers on inference surfaces

  #1609·B-A-M-N opened Oct 20, 2026Contributor·

  ·
  1

  3

• fix(routing): stop counting key-pool exhaustion against the breaker

  #1607·B-A-M-N opened Oct 20, 2026Contributor·

  ·
  1

• fix(serving): enforce the model tier gate on /v1/embeddings

  #1602·B-A-M-N opened Oct 19, 2026Contributor·

  ·
  7

• fix(serving): bound the forced-stream hand-off queue and make reader cancellation safe

  #1560·B-A-M-N opened Oct 19, 2026Contributor·

  ·
  2

• fix(serving): bound the request-log queue and drain it on shutdown (#133)

  #1559·B-A-M-N opened Oct 19, 2026Contributor·

  ·
  1

  1

• fix(serving): keep unknown cache usage distinct from a measured miss

  #1517·B-A-M-N opened Oct 11, 2026Contributor·

  ·
  48

• feat(batch): add general batch processing endpoint (WIP)

  #1510·prdai opened Oct 10, 2026·

  ·
  17

• fix(serving): require evidence of model progress before clearing serving health

  #1502·B-A-M-N opened Oct 4, 2026Contributor·

  ·
  44

• fix(routing): stop transient HTTP-200 responses from poisoning prefix-locality evidence

  #1501·B-A-M-N opened Oct 4, 2026Contributor·

  ·
  24

• fix(auth): bound login refusals without growing attempt queues

  #1468·B-A-M-N opened Sep 23, 2026Contributor·

  ·
  17

• feat(storage): isolate hard-delete claim state

  #1463·B-A-M-N opened Sep 22, 2026Contributor·

  ·
  28

• fix(routewise): propagate cold-prefill scheduling priority

  #1461·B-A-M-N opened Sep 21, 2026Contributor·

  ·
  14