v0.6.3

What's new in 0.6.3 (2026-07-24)

These are the changes in xagent v0.6.3.

New features

• feat: replace fixed skill/memory pipeline steps with model-driven tools by @qinxuye in #917
• feat: unify page top bars and redesign template cards by @bluefish-08 in #923
• feat: add native video understanding and durable file refs by @qinxuye in #924
• feat: add SEO Brief Writer, Research & Enricher, KB Writer and Email Lead Response templates by @yiboyasss in #934
• feat: add Google Docs, Google Slides and HubSpot MCP connectors by @yiboyasss in #938
• feat: allow enabling triggers during agent creation (#928) by @codeacme17 in #937
• feat(workforce): add runs history for workforce (#801) by @codeacme17 in #933
• feat(workforce): remove redundant "Workforce Instruction" field (#800) by @codeacme17 in #939
• feat: inspect Workforce agent executions by @qinxuye in #919
• feat(agent): hide "Agent Invocation" tool from agent configuration (#802) by @codeacme17 in #940
• feat(workforce): deployment foundation — deployments table, run source & idempotency, archive termination, config fingerprint by @codeacme17 in #952
• feat(models): split OpenAI provider into official OpenAI and OpenAI-Compatible by @kejhz653-stack in #959
• feat: add workforce shareable link deployment by @codeacme17 in #967
• feat(workforce): webhook/trigger deployment channel (#950) by @codeacme17 in #966
• feat: add Workforce REST API and SDK deployment channel by @codeacme17 in #968
• feat(workforce): widget deployment by @codeacme17 in #985
• feat: add personal API key management by @AlexLiu190625 in #970
• feat: managed SSH MCP tools with sandboxed execution by @bluefish-08 in #990

Enhancements

• enh: track provider prompt-cache token usage by @qinxuye in #922
• enh: parallelize KB web ingestion (page + batch concurrency) to fix Step 2 slowdown by @bluefish-08 in #915
• enh: compact task channel labels by @qinxuye in #962
• enh: update get started card links to docs.xagent.co subpages by @yiboyasss in #974
• enh: improve branded static visual generation by @qinxuye in #942
• ref(memory): hoist shared filter dispatch into MemoryStore base + None date-bound guard + top-k post-filter docs (#916) by @codeacme17 in #918

Bug fixes

• fix(i18n): translate login hero tagline to Chinese in zh locale by @yiboyasss in #914
• fix: drop redundant worker liveness pre-check when enqueuing KB web ingest jobs by @bluefish-08 in #913
• fix(kb): parse staged background uploads instead of the unpublished canonical path by @bluefish-08 in #932
• fix: cancel in-flight tool executions by @qinxuye in #921
• fix(migrations): merge parallel 2026-07-20/21 alembic heads by @codeacme17 in #941
• fix: harden agent execution reliability by @qinxuye in #943
• fix(triggers): exclude workforce-generated manager agents from trigger ownership resolution by @codeacme17 in #951
• fix: handle Workforce dependencies during Agent deletion by @AlexLiu190625 in #954
• fix: persist omitted tool_categories as NULL instead of [] (#944) by @codeacme17 in #957
• fix(websocket): harden connection lifecycle by @AlexLiu190625 in #963
• fix: prevent structured agent update errors from crashing the builder by @AlexLiu190625 in #961
• fix(widget): re-validate agent widget guest tokens per request by @codeacme17 in #989

Others

• refactor(tools): prefetch runtime database state with worker-owned sessions by @AlexLiu190625 in #964
• chore: bump compose images to 0.6.2 by @qinxuye in #920

Full Changelog: v0.6.2...v0.6.3