DEV Community

Fenju Fu
Fenju Fu

Posted on

Astron Open Source Β· Bi-Weekly Report (Jun 29 – Jul 12, 2026)

#agents #ai #news #opensource

🌟 Highlights

  • Holding strong on the OSS Insight Java trending board: across the last three months of Java repositories, SkillHub #4 Β· Astron Agent #5 β€” both firmly in the Top 5.
  • Two new projects open-sourced (Jul 2): MemFlywheel β€” a file-native long-term memory layer for Agent Harnesses that makes agents smarter with every run; Domux β€” a low-latency semantic-parsing model fine-tuned from Gemma-4 that turns natural-language home commands into structured data within 150 ms (98.37% accuracy).
  • A double release: Astron Agent v1.1.0 (Jul 6) gives the Standard Agent Workflow invocation + Link plugins + Mem0 long-term memory in one release; SkillHub v0.2.13 (Jul 3) polishes review governance, password security, and SUPER_ADMIN protection.
  • Zero-friction onboarding + two offline events: the Astron Agent open-source edition online playground went live (Jun 30 β€” try it in the browser, no deployment); HER Hack-Astron #1, the wearables Γ— emotional-companion agent challenge (Jul 6), and the 2nd AI Innovation & Development Conference in Tianjin (Jul 11, with Dong Jiang presenting "Enterprise Agentic AI Application Architecture & Knowledge Engineering") landed back-to-back.

OSS Insight Β· Java 3-month trending board

πŸ‘₯ Top 10 Active Community Contributors (Jun 29 – Jul 12, 2026)

Based on message volume in the open-source discussion group (automated accounts such as the Astron Q&A bot and the welcome bot excluded); nicknames anonymized. Thank you to everyone who answered questions, asked questions, and built alongside us πŸ™Œ

Astron community Top 10 active contributors

πŸ“¦ Release Highlights

πŸ€– Astron Agent v1.1.0 (Jul 6) β€” Workflow + Plugins + Long-Term Memory, all in one

  • πŸš€ Workflow invocation inside Agents: Standard Agents can persist workflow capabilities and expose them as runtime tool callbacks, usable from both chat and debug paths, with ownership and space-membership validation (#1487).
  • ✨ Link plugin capability: import Link plugins from the plugin store on the ability page, persist selected tools, and invoke them directly in debug/chat; official/admin-owned plugins are treated as trusted import sources (#1448).
  • 🧠 Mem0 long-term memory: configurable Mem0 memory with app-level isolation, non-blocking writes, and improved Chinese memory preservation and retrieval (#1479).
  • πŸ”’ Hardened official tool debug-URL validation and trusted-owner handling (#1453); fixed a GitHub code-scanning SSRF alert (#1459); several rounds of CI/remote-test stabilization.
  • πŸ”§ Upgrade notes: deploy Console/Toolkit changes together to enable workflow invocation; review plugin/workflow/Mem0 settings on existing agents after upgrading. Change scope: 154 files, +6,489/βˆ’6,559 lines.
  • πŸ“ Release: https://github.com/iflytek/astron-agent/releases/tag/v1.1.0

πŸ›‘οΈ SkillHub v0.2.13 (Jul 3) β€” Review governance & permission hardening

  • ✨ Promotion review upgrade: pending review cards show richer metadata, with status/reviewed-time sorting and expanded backend/frontend + Playwright coverage (#555).
  • πŸ”‘ Password settings restored: a backend-authoritative canChangePassword capability β€” local-password users can change passwords again, while OAuth-only accounts show an unavailable state (#554).
  • πŸ”’ SUPER_ADMIN hardening: non-super-admin users can no longer assign or tamper with the SUPER_ADMIN role (#568).
  • πŸ› Fixed scanner local LLM Base URL propagation while keeping the scan API contract stable, with Kubernetes docs and regression coverage (#565); nested relative links now work in skill-package markdown previews (#557); patched the vulnerable undici frontend dependency (#567).
  • πŸ“ Release: https://github.com/iflytek/skillhub/releases/tag/v0.2.13 ｜ No breaking changes β€” safe to upgrade.

🧩 Newly open-sourced: MemFlywheel & Domux (Jul 2)

  • 🧠 MemFlywheel: a file-native long-term memory layer for Agent Harnesses β€” pre-run recall, post-run extraction, and skill distillation; memories are stored as Markdown, so they are inspectable, Git-diffable, and fully evidence-linked. Works with Pi, Hermes, OpenCode, and other mainstream harnesses β†’ https://github.com/iflytek/memflywheel
  • ⚑ Domux: a low-latency semantic-parsing model fine-tuned from Gemma-4-E2B-it (SFT + GRPO) β€” parses natural-language smart-home commands into a 7-field structured schema within 150 ms end-to-end, at 98.37% accuracy and 100% format compliance, with no device whitelist required; built for edge deployment β†’ https://github.com/iflytek/domux

βš™οΈ Astron RPA β€” Scheduled-task trigger support special

  • πŸ› A user reported nightly scheduled tasks occasionally not firing: the team traced it in the logs to a too-short trigger grace window (a busy machine can miss the 1-second window), shipped a hotfix on the spot (patched trigger.py / base_task.py with a wider grace period + verbose logging), and recommended advanced cron expressions; the fix will be merged back upstream.
  • πŸ“š Added an AGENTS.md project-context document (#813), aligning with the AI coding-agent ecosystem standard.

πŸ’¬ Community Q&A & Discussions (Bi-Weekly Picks)

Grouped by product line; all are real community questions + the engineering team's solutions. 🟒 Resolved / 🟑 In progress.

SkillHub (mostly ecosystem integration & governance)

  • 🟑 Could an MCP marketplace be integrated into SkillHub to build a unified enterprise capability platform (Skills + MCP)? Architecturally feasible β€” SkillHub uses pluggable SPI boundaries and already ships a ClawHub CLI compatibility layer as a foundation; logged as a product-direction request.
  • 🟑 When an agent improves a skill, can it contribute the improvement back to the shared registry as a PR? A great idea for closing the skill co-building loop; the user was invited to file an issue.
  • 🟒 v0.2.13 upgrade notes: no breaking changes; if your Scanner uses a local LLM, mind the Base URL setting (the propagation bug is fixed in this release).

Astron Agent (mostly second-development & new capabilities)

  • 🟒 White-labeling for enterprise redistribution: after changing frontend branding, rebuild the image with each module's Dockerfile under its build directory and swap the official image (same for backend); along the way the team also resolved a Docker image pull failure (switch to a domestic mirror) and a workflow publish failure (fill in the workflow's basic info at the top-left of the canvas first) β€” the customer demoed successfully the next day.
  • 🟒 Is the platform decoupled from iFLYTEK models? Long since β€” any OpenAI-protocol-compatible third-party or self-hosted model can be added on the Model Management page.
  • 🟒 Is there an "autonomous planning" agent capability? The platform already supports multi-agent orchestration, workflows, tool/MCP integration, and Agentic Workflow Auto-Generation; file an issue for a standalone autonomous-planning app.
  • 🟒 An intranet WPS document-processing plugin? Not yet β€” issues/PRs welcome.

Astron RPA (mostly deployment environments & builds)

  • 🟒 Nightly scheduled task not firing: see the special above β€” grace-window hotfix + verbose-log diagnosis, closed the same day.
  • 🟒 Does it run on cloud desktops / VMs? Yes (the team routinely tests on Tencent cloud desktops); requires Win10 1809 or later; one Windows Server login-hang case is being remote-diagnosed by support 🟑.
  • 🟒 Manual client build fails with Cannot compute electron version: usually pnpm missing from PATH / frontend dependencies not installed β€” check the environment against BUILD_GUIDE.zh.md (no Java/Docker needed); remote assistance offered.
  • 🟒 How do teams keep script versions consistent without git-style merging? Use the built-in workflow version management (publish/rollback/diff) plus the Center of Excellence as the team's centralized "process market" distributing stable versions.
  • 🟒 Can the open-source RPA be invoked standalone, without the Agent platform? Yes; the cloud RPA has no standalone API yet β€” call it through the Agent platform and expose the agent as an API, with a video course and a hands-on article to follow.

πŸ“Œ Q&A entry points: @-mention the Astron Q&A bot in the group (24/7), or ask on the web β†’ https://awesome-astron-workflow.dev/chat

πŸ™Œ Community Call: Come Contribute! (Good First Issues & Bounties)

Newcomer-friendly and rewarding from your first commit (contribution points are redeemable at the points store):

🌱 Good First Issues (great for a first contribution)

Project Issue What it's about
Astron Agent #1412 Expand the docs FAQ
Astron Agent #1410 Contribute a workflow example to the gallery
Astron Agent #1405 "Are you using Astron Agent?" user-story collection
Astron RPA #805 "Are you using Astron RPA?"
SkillHub #528 "Are you using SkillHub?"

πŸ’° Bounties (reward Β· more challenging, more rewarding)

Project Issue What it's about
Astron Agent #1067 Let OpenClaw trigger Astron Agent to schedule an RPA bot for secure operations
Astron Agent #1066 Build a "ChatClaw" app on the drag-and-drop canvas with visual fine-tuning
Astron Agent #709 Support the A2A (agent-to-agent) protocol

πŸ‘‰ Want to claim one? Just comment on the issue. Not sure where to start? Ping us in the discussion group and the engineering team will help break the task down.

✨ Community Spotlight

  • 🌐 Astron Agent open-source online playground is live (Jun 30): zero deployment, zero configuration β€” try it right in the browser β†’ https://astron-agent-nginx.zeabur.app/
  • πŸ’œ HER Hack-Astron #1｜Wearables Γ— Emotional-Companion Agent Challenge (Jul 6): a dozen-plus teams built agents on site β€” an AI sleep companion, a health guardian, and "Heart Companion," a retro pocket-watch-style emotional-companion agent (the first submission PR, #1493); see issue #1482 for the event archive.
  • 🎀 2nd AI Innovation & Development Conference, Tianjin (Jul 11): iFLYTEK senior architect Dong Jiang presented "Enterprise Agentic AI Application Architecture & Knowledge Engineering"; the community offered a 15-point ticket-redemption perk.

πŸ’¬ Got questions, ideas, or want to contribute? Join the discussion group, or head straight to the repos under iflytek-astron to open an Issue / PR.
Let's build a thick, solid open-source ecosystem for enterprise-grade AI Agents together πŸš€

Originally published in the iFLYTEK open source community discussions: https://github.com/orgs/iflytek/discussions/6

Top comments (0)