humpyard v0.15.0

What's Changed

• feat(routing): a rejected request moves a pool on, within a bound (ADR 0026) by @gschaetz in #183
• Release 0.15.0 by @gschaetz in #184

Full Changelog: v0.14.1...v0.15.0

humpyard v0.14.1

What's Changed

• Record the first week of quota measurements (task 2.1 stays open) by @gschaetz in #178
• ctl stats: say an empty result is scoped to the key by @gschaetz in #180
• feat(config): provider tool_call_extra_content, for Gemini's thought signature by @gschaetz in #181
• Release 0.14.1 by @gschaetz in #182

Full Changelog: v0.14.0...v0.14.1

humpyard v0.14.0

What's Changed

• feat(config): provider drop_fields, and read array-shaped provider errors by @gschaetz in #177
• Release 0.14.0 by @gschaetz in #179

Full Changelog: v0.13.0...v0.14.0

humpyard v0.13.0

Highlights

Expiring runtime overrides. Take a target out of rotation, drain a provider, or pin a route to one target, right now, with no config edit and no reload. Every override expires (4 hours by default, 24 at most), is audited with the creating key's id, and can only remove options, never add them. If an override would leave a route with nothing to use, the request fails with a clear 503 instead of going somewhere you did not choose.

humpyard ctl overrides add --disable-target deep --reason "slow since 02:00" --for 2h
humpyard ctl overrides add --drain-provider groq --reason "rate limited"
humpyard ctl overrides add --pin auto=fast --reason "keep it local" --for 30m
humpyard ctl overrides list [--history]
humpyard ctl overrides remove 3

Overrides show up in decisions, explain, GET /v1/config and two metrics, with an alert (HumpyardOverrideLong) for one left in force too long. They need a ledger and a key with manage_overrides = true. See Runtime overrides.

Also new

• humpyard init --out - prints the config to standard output (progress and the one-time key go to standard error), so it can be piped, for example into kubectl create configmap. (#160, @LuckTerence)
• humpyard completions <shell> for bash, elvish, fish, PowerShell and zsh. (#161, @LuckTerence)
• tokio updated to 1.54.1 for RUSTSEC-2026-0386.

Upgrading

The ledger moves to schema v10 the first time 0.13.0 starts on an existing database. Existing decisions and usage are kept. An older binary cannot open a v10 ledger, so copy the ledger first if you may want to go back.

Override state is per instance: with several gateways, apply an override to each one.

Pre-alpha. Apache-2.0.

---

What's Changed

• Spec: expiring runtime overrides (disable a target, drain a provider, pin a route) by @gschaetz in #159
• Add humpyard init --out - to print config to stdout by @LuckTerence in #160
• Spec: init --out - prints the configuration to standard output by @gschaetz in #162
• Ask for a star in the README header by @gschaetz in #167
• Refresh the roadmap and architecture status by @gschaetz in #168
• Runtime overrides, part 1: settings, key setting and the store by @gschaetz in #169
• Runtime overrides, part 2a: the API and the planning step by @gschaetz in #170
• Runtime overrides, part 2b: drain_provider by @gschaetz in #171
• Runtime overrides, part 3: decisions, explain, config view and metrics by @gschaetz in #172
• Runtime overrides, part 4a: humpyard ctl overrides by @gschaetz in #173
• Runtime overrides, part 4b: property and architecture tests; archive the change by @gschaetz in #174
• Bump tokio to 1.54.1 (RUSTSEC-2026-0386) by @gschaetz in #175
• feat(cli): add humpyard completions for bash, zsh and fish by @LuckTerence in #161
• Release 0.13.0 by @gschaetz in #176

New Contributors

• @LuckTerence made their first contribution in #160

Full Changelog: v0.12.0...v0.13.0

humpyard v0.12.0

What's Changed

• Update rustls-webpki and webpki-root-certs for RUSTSEC-2026-0382 by @gschaetz in #153
• Add the effective configuration: GET /v1/config, ctl config, and the fingerprint by @gschaetz in #152
• Release 0.12.0: the effective configuration, and the TLS advisory fix by @gschaetz in #154

Full Changelog: v0.11.0...v0.12.0

humpyard v0.11.0

What's Changed

• Add humpyard ctl: a read-only client for a running gateway by @gschaetz in #149
• A 413 is overflow: move on without counting it against the endpoint by @gschaetz in #150
• Release 0.11.0: humpyard ctl, and a 413 is overflow by @gschaetz in #151

Full Changelog: v0.10.0...v0.11.0

humpyard v0.10.0

What's Changed

• Define the UI configuration model (ADR 0025) by @gschaetz in #140
• Measure ledger scale and gateway overhead (performance.md) by @gschaetz in #141
• Docs: why session stickiness is not planned by @gschaetz in #143
• Spec: per-request cost ceiling on keys (designed, not built) by @gschaetz in #142
• Fix the quick start: Groq no longer serves its example model by @gschaetz in #144
• Spec: a guided humpyard init that checks models live by @gschaetz in #145
• Add humpyard init: a guided first config that checks models live by @gschaetz in #146
• Spec: a read-only humpyard ctl client for a running gateway by @gschaetz in #147
• Release 0.10.0: humpyard init, a guided first config by @gschaetz in #148

Full Changelog: v0.9.5...v0.10.0

humpyard v0.9.5

What's Changed

• Treat 410 and 402 as unavailable endpoints by @gschaetz in #138
• Release 0.9.5 by @gschaetz in #139

Full Changelog: v0.9.4...v0.9.5

humpyard v0.9.4

What's Changed

• Redraw the request flow diagram top to bottom by @gschaetz in #133
• Add --rename and --quality-weight to migrate-modelrelay by @gschaetz in #135
• Guarded tail for pool routes (add-tiered-fallback) by @gschaetz in #136
• Release 0.9.4 by @gschaetz in #137

Full Changelog: v0.9.3...v0.9.4

humpyard v0.9.3

What's Changed

• Add quality-aware ordering (add-quality-ranking) by @gschaetz in #131
• Archive the finished OpenSpec changes by @gschaetz in #132
• Release 0.9.3 by @gschaetz in #134

Full Changelog: v0.9.2...v0.9.3