v0.10.17 - The Stars

fast-agent 0.10.17

• Expanded model support: GPT-6 Astra, Muse Spark 1.3, Gemini 3.8 Flash, Claude Fable 5.1, GLM-5.3/Flash and DeepSeek V4 Flash Vision.
• Astra defaults: medium reasoning across the picker, CLI and codexplan, plus opt-in extended context.
• Codex web tools: standalone search and browsing, with search references preserved across resumed conversations.
• Better process management: automatic reattachment on session resume, /process terminate and clearer status reporting.
• Provider and runtime improvements: Codex fast mode, Bedrock reasoning controls, more reliable reasoning replay, streaming and context compaction.
• Fixes and hardening: marketplace path containment, tool-result preservation at iteration limits, MCP OAuth compatibility, macOS Java bootstrap and more accurate trace usage accounting.

Thanks to new contributors!

What's Changed

• fix(cli): write scaffold files as UTF-8 by @panDing19 in #931
• glm flash, ux process/socket tweaks by @evalstate in #932
• Dev/0.10.12 by @evalstate in #933
• Add DeepSeek V4 Flash Vision support by @evalstate in #934
• Improve test performance, Bedrock support, and Codex fast mode by @evalstate in #936
• fix(llm): omit Glimmer default max tokens by @evalstate in #937
• fix: preserve tool results when iteration limit is reached by @oodadoudou in #940
• fix(marketplace): contain skill and plugin installs by @Chan-Developer in #939
• fix: resolve macOS Java home for CPD bootstrap by @oodadoudou in #942
• test: add generate coverage for ChainAgent cumulative mode by @Christian-Sidak in #935
• Release 0.10.17 by @evalstate in #943

New Contributors

• @oodadoudou made their first contribution in #940
• @Chan-Developer made their first contribution in #939
• @Christian-Sidak made their first contribution in #935

Full Changelog: v0.10.10...v0.10.17