0.4.5

[2026-08-20] The playground panel groups its fields and starts in debug mode. (details, #181)
[2026-08-20] Clients accept default headers for endpoints that demand their own. (details, #181)
[2026-08-20] AutoLLMClient lists the model ids the configured endpoint serves, filtered to the routed client, and the playground lists them into its model dropdown. (details, #181)
[2026-08-20] The Responses-protocol message transforms match the GPT client, including when buffered text is flushed. (details, #181)
[2026-08-20] Unknown stream output is skipped unless AGENTHUB_DEBUG is set. (details, #181)
[2026-08-19] OpenRouter registry entries move from the openai-chat client to openai-responses. (details, #180)
[2026-08-19] The live E2E suites retry rate-limited responses instead of failing the run. (details, #180)
[2026-08-19] claude-opus-5 joins the registry, and the official GPT-5.6 entry is named gpt-5.6-sol. (details, #180)
[2026-08-19] The playground model dropdown follows the E2E model list, and an option can declare its client type. (details, #180)
[2026-08-19] GLM-5.3 launched: the docs snapshot, registry pricing, and E2E model move to the live API. (details, #180)

Navigation Menu