Fixed
- (deps) Update module github.com/mark3labs/mcp-go to v1.1.1 in #498 by @renovate[bot]
Full Changelog: https://github.com/giantswarm/klaus/compare/v0.1.30...v0.1.31
AI Agents
Fixed
- (deps) Update module github.com/giantswarm/klaus-oci to v0.0.76 in #220 by @renovate[bot]
Full Changelog: https://github.com/giantswarm/klaus-operator/compare/v0.0.168...v0.0.169
Changed
- Align files according to platform standards in #655 by @giantswarm-align-files[bot]
Full Changelog: https://github.com/giantswarm/agent-platform/compare/v4.66.2...v4.66.3
Fixed
- The model-manager floor is
1.0.8: the Ollama ModelConfigs it wires carrynum_ctx, so agent prompts are no longer cut on hosts with less than 24 GiB of VRAM (giantswarm/model-manager#157). Ollama picks its default context window from the host’s VRAM: 4,096 tokens below 24 GiB, which is every CPU-only host. It fills a longer prompt by dropping everything between the first tokens and the tail, and an agent turn loses its system prompt and tool schemas without an error. model-manager 1.0.8 writesspec.ollama.options.num_ctxon every Ollama ModelConfig it creates or re-wires: its chart valueollama.contextLength, default 32768, capped at the model’s own context length. Loads pre-warm at the same window.components.model-manager.versionRangeis>=1.0.8 <2.0.0and the BOM pins1.0.8. The previous range already admitted 1.0.8; the floor keeps an installation from resolving an older release. Installations without an Ollama backend see no change.
- The model-manager floor is
Changed
- muster’s floor is 5.31.4:
components.muster.versionRange>=5.31.4 <6.0.0, the example customer BOM pins5.31.4(giantswarm/muster#1323). A session on a forwarded bearer — an agent’s on-behalf-of token, a forwarded ID token — ends at the bearer’sexp: its backend connections close, and a server whose last connection it held goes back toAwaiting Session. Before, the session’s pooled connections kept forwarding the expired token on every capability poll and listen-stream reconnect, each backend answered 401, and only the third rejection in a row evicted a connection, with a WARN for every attempt. The expected provider-token store miss of such a session is DEBUG instead of about 30 WARN lines per session start. The values the chart forwards do not change; an installation on the 5.x line rolls muster once.make verify-presetsrenders the presets through the floor.
- muster’s floor is 5.31.4:
Changed
- Align files according to platform standards in #1328 by @giantswarm-align-files[bot]
Full Changelog: https://github.com/giantswarm/muster/compare/v5.31.4...v5.31.5
Changed
- Align files according to platform standards in #221 by @giantswarm-align-files[bot]
Full Changelog: https://github.com/giantswarm/klaus-operator/compare/v0.0.167...v0.0.168
Changed
- Align files according to platform standards in #500 by @giantswarm-align-files[bot]
Full Changelog: https://github.com/giantswarm/klaus/compare/v0.1.29...v0.1.30
Changed
- klaus-gateway’s floor is 3.3.0:
components.klaus-gateway.versionRange>=3.3.0 <4.0.0, the example customer BOM pins3.3.0(giantswarm/klaus-gateway#329). A Slack turn that fails before the agent answered says what broke — the agent’s tools, the platform connection, the model, a policy — instead of always asking the person to try again, a tool-set or platform failure is sent once more on the thread’s AgentInstance first, andturn_completeandklaus_gateway_turn_totalcarryfailure_class, so a burst of tool-set failures can alert apart from a one-off model error. The values the chart forwards do not change; an installation on the 3.x line rolls the gateway once.
- klaus-gateway’s floor is 3.3.0:
Changed
- Align files according to platform standards in #91 by @giantswarm-align-files[bot]
Full Changelog: https://github.com/giantswarm/klaus-oci/compare/v0.0.77...v0.0.78