v0.9.13

0.9.13 (2026-07-29)

Features

• chart: template Grafana dashboards, publishing runtime ones only while their runtime serves (#1267) (c2d56b7)
• foreman: test-dilution advisory for the coder gate (#1332) (#1337) (43a30fc)

Bug Fixes

• controller: append TGI extraArgs once, not four times (#1306) (6d675f2)
• controller: Available goes False when no replicas are ready (#1304) (f3bd243)
• controller: charge GPUQuota by the resolved GPU count, closing the Model-declared-count bypass (#1324) (e052229)
• controller: cpu-offload verify advisory, parallelism validation, and a sample that actually runs (#1321) (9740dad)
• controller: gate HPA scale-up by GPUQuota via maxReplicas accounting (#1311) (#1334) (dcefc45)
• controller: recognise huggingface.co URL sources as HF repos, not single-file downloads (#1322) (#1325) (36fc79f)
• controller: revalidate model cache via HEAD + Content-Length, atomic download (#1309) (#1326) (3e86988)
• foreman: accept multi-segment repo slugs (GitLab subgroups, nested Forgejo orgs) (#1299) (#1341) (1b7c47a)
• foreman: auto-apply gofmt in the coder gate instead of failing the coder (#1327) (#1333) (954c67b)
• foreman: do not downgrade a mixed footprint whose classes are all self-GO (#1342) (#1344) (d7c6569)
• foreman: make coder self-gate diff-aware, skip Go tier on non-Go diffs (#1292) (#1336) (9f4acef)
• foreman: ParseRepo delegates to the shared codehost slug validator (#1343) (#1345) (ae2dfc2)
• foreman: record IsError on MCP transport failures (#1330) (#1335) (c42f873)
• foreman: split git committer identity from author for human DCO sign-off (#1323) (44ef09f)
• foreman: test-presence accepts behavioral coverage of unexported helpers (#1329) (#1340) (f1aea17)

Documentation

• api: correct MemoryLimitGiB doc, it is quota-accounting only with no runtime enforcement (#1312) (#1339) (90e6ca3)
• api: stop the GPU-memory fields implying enforcement that does not exist (#1314) (8660c85)
• foreman: replace em dash with colon in README heading (#1296) (8586a37)
• multi-tenancy: document GPUQuota is off by default + opt-in/Kueue/HPA/placement notes (#1313) (#1338) (3081897)