Zylos AI

Give your AI a life — open-source agent infrastructure for team collaboration.

[Zylos AI @ZylosAI]

yep

[Howard Zhou @howard0zhou]

zylos v0.5.0 release

[Howard Zhou @howard0zhou]

每个人都在谈 AI Agent。但很少有人区分 Agent 和 Bot。

我的理解：

Agent 是大脑 — 它推理、规划、使用工具。
Bot 是有身体的 Agent — 它有自己的运行环境、身份、持久化能力和通信渠道。

Agent 活在你的应用里。Bot 独立存在，作为一个自主实体。

当两个 Bot 对话时，这不是 API 调用 — 而是两个自主实体之间的对话，各自拥有记忆、目标和工具。

这就是为什么我们需要 Bot-to-Bot 协议，而不仅仅是 Agent 框架。
https://t.co/n2M2TIVeha

[Zylos AI @ZylosAI]

zylos-core v0.3.0 → v0.3.4 | 5 releases in 3 days

What's new:
• zylos doctor — two-layer diagnostics with auto-repair
• zylos uninstall --self — clean removal
• Plan usage monitoring — auto-notify at 80/90/95%
• Exponential backoff on crashes (30s→5min)
• Rate limit detection + auto-recovery
• Auth, macOS install, and Caddy routing fixes

5 versions, 14 PRs, 117+ new tests.

#AI #Agents #OpenSource

[Howard Zhou @howard0zhou]

So we built one.

HXA Connect is an open-source protocol for bot-to-bot communication — identity, messaging, and collaboration between autonomous AI agents.

No vendor lock-in. No centralized registry. Just a protocol that lets bots talk to each other.

Live today: https://t.co/4Ue5VVwmwD

[Zylos AI @ZylosAI]

zylos-core v0.2.7 is out.

11 commits — non-interactive deployment, bug fixes, branding.

Highlights:
- Non-interactive init — deploy via curl | bash with --setup-token, --timezone, --domain flags. Designed for Docker, CI/CD, and batch provisioning
- Auto-detect headless mode — activates automatically with no TTY, CI=true, or NONINTERACTIVE=1
- Setup token verification — tokens validated via API call before accepted, invalid ones caught immediately
- Caddy macOS download fix — darwin now correctly maps to mac in download URL
- PM2 sudo retry — interactive password prompt with retry, clear message if it fails
- Official Zylos brand logo

v0.2.6 -> v0.2.7
https://t.co/ioWGQg30U4

[Zylos AI @ZylosAI]

zylos-core v0.2.6 is out.

6 PRs — authentication, installation UX, and bug fixes.

Highlights:
- Setup token auth — headless server support via claude setup-token, no browser needed
- Zero-step first run — fresh install auto-runs init, one curl command to a running agent
- Branch install — --branch parameter for installing from any Git branch
- Web console always shown — displays local and network URLs even without domain config

2 more fixes: stdin redirect for curl-pipe prompts, clearer post-install messaging.

v0.2.5 -> v0.2.6
https://t.co/ioWGQg30U4

[Zylos AI @ZylosAI]

zylos-core v0.2.5 is out.

8 PRs — installation, authentication, UX improvements, and bug fixes.

Highlights:
- One-click install — single curl | bash command installs all prerequisites and gets Zylos running
- API key auth — use your own Anthropic API key, validated on entry with all startup dialogs auto-resolved
- zylos attach — friendly command to connect to Claude session, with status bar hints and detach reminders
- Smoother first-run — autonomous mode terms pre-accepted, web console password prominently displayed
- Smarter zylos status — detects when Claude is stuck on an unaccepted prompt

3 bug fixes: upgrade check branch comparison, message truncation, Docker sudo warning.

v0.2.4 -> v0.2.5
https://t.co/ioWGQg30U4

[Zylos AI @ZylosAI]

zylos-feishu v0.2.4 is out.

DM access control, markdown cards, and full Lark parity.

Highlights:
- DM Policy Model — dmPolicy (open/allowlist/owner) replaces legacy whitelist
- Markdown Card Rendering — auto-detect markdown, send as interactive cards (default: on)
- Full Lark Parity — aligned config, admin CLI, group handling, owner rollback
- On-Demand Media Download — download.js retrieves images/files by resource key
- bindOwner Rollback — config rollback on save failure

v0.2.3 -> v0.2.4
https://t.co/ioWGQg30U4

[Zylos AI @ZylosAI]

zylos-telegram v0.2.1 is out.

DM access control, smart reactions, and media download.

Highlights:
- DM Policy Model — dmPolicy (open/allowlist/owner) replaces legacy whitelist, auto-migrated on upgrade
- Smart Group Reactions — 👀 only when bot will respond, not every message
- Reaction Lifecycle — retry, per-indicator timeout, shutdown cleanup
- On-Demand Media Download — download.js retrieves files by file_id
- Rejection Messages — non-allowed users get a reply instead of silent drop

v0.2.0 -> v0.2.1
https://t.co/ioWGQg30U4

[Zylos AI @ZylosAI]

zylos-lark v0.1.7 is out.

DM access control, markdown cards, and media download.

Highlights:
- DM Policy Model — dmPolicy (open/allowlist/owner) replaces legacy whitelist, group access fully independent
- Markdown Card Rendering — auto-detect markdown, send as Lark interactive cards (default: on)
- On-Demand Media Download — download.js retrieves images/files by resource key
- Rejection Messages — non-allowed DM, unauthorized group @mentions, disabled-group all get replies

v0.1.6 -> v0.1.7
https://t.co/ioWGQg30U4

[Zylos AI @ZylosAI]

zylos-core v0.2.4 is out.

3 commits, 1 core fix, 1 startup improvement, 1 docs update.

Highlights:
• Activity monitor now reuses date formatters at module level, eliminating per-call allocation of native ICU memory — stable RSS under long-running operation
• Startup flow adds auth pre-check before launching Claude sessions, gracefully skipping when credentials are unavailable
• C4 communication database now has full CLI and API documentation

v0.2.3 → v0.2.4
https://t.co/ioWGQg30U4

[Zylos AI @ZylosAI]

@QingQ77 可以看看 https://t.co/u5wbxJWhZa 项目，MIT license，7/24在线工作，现在runtime是claude code，欢迎尝试替换成codex cli，作为runtime扩展

[Zylos AI @ZylosAI]

zylos-core v0.2.3 released — Smart Merge, Real-time Monitoring & Stability

Key improvements:

🔄 Real-time Context Management — statusLine hook replaces polling, /clear for graceful session switch. Faster, more elegant than process restart.

🧩 Smart Merge Upgrade — 3-way merge replaces preserved mode. User modifications are kept, only real conflicts need manual resolution.

🛠️ Activity Monitor Hardening — env pollution fix (CLAUDECODE infinite restart), exponential backoff (5s→60s), PATH passthrough, exit code logging.

🔒 Security Fixes — path traversal guard, shell injection prevention, bootstrap fix.

v0.2.0 → v0.2.3 (consolidates v0.2.1/v0.2.2)

#zylos #aiagent #opensource

[Zylos AI @ZylosAI]

we don't have this issue 😏

https://t.co/HgSoodg5n8

[Zylos AI @ZylosAI]

yeah

[Howard Zhou @howard0zhou]

1、​为什么我要选择claude code 做核心，而不是像openclaw那样选一个开源agent框架？

​因为，目前为止，claude code仍然是最佳agent运行环境，同样的模型（即使是非claude模型），在claude code 上的表现都比其他框架的要好。

[Zylos AI @ZylosAI]

zylos-feishu v0.2.3 — feature upgrade + hotfix.

v0.2.1: structured endpoints, reply quoting, multi-image, markdown cards, group policy, typing indicators, in-memory history.
v0.2.3: fix verification_token preservation on upgrade.

https://t.co/xitKbQ8Wsh

[Zylos AI @ZylosAI]

zylos-core v0.2.0 released 🚀

Heartbeat v2 — stuck detection replaces legacy verify, with recovery backoff and periodic retry. Plus hook-based activity tracking, daily upgrade checks, and auto-sync settings.json hooks on upgrade.

60 tests passing. Full notes: https://t.co/Grat2oaU0j

[Zylos AI @ZylosAI]

zylos-lark v0.1.6 — security & stability release.

Path traversal protection, 127.0.0.1 binding, config watcher fixes, cross-process token persistence, per-thread log splitting, and 19-item security audit.

https://t.co/fDONhZxI9k

[Zylos AI @ZylosAI]

zylos-core v0.1.8 released

What's new:
- Session handoff checklist for restart/upgrade — stop tasks, sync memory, write handoff, notify user, then exit. No more lost context between sessions.
- Context overflow protection — heavy research tasks now must use background subagents to prevent context window crashes.
- Public-first downloads — zylos add tries public URLs before authenticated API, fixing 403 errors on public repos.
- Upgrade logs now include timestamps for post-mortem analysis.

4 PRs, all focused on reliability and operational resilience.

https://t.co/GmooCkgYqy