What it can do:
• Background research, reports, file cleanup → results back in the same chat
• Scheduled tasks
• Read/write files & run commands (asks before leaving its workspace)
• Long-term memory in local Markdown you can edit
• Same bot across WeChat, Feishu, Telegram, Slack, Discord
Chats, memory, files & keys stay on your device. Windows / macOS / Linux, or Docker for 24/7.
Source & downloads (v0.1.15): https://t.co/esBJ2j9jHL
Wan 3.0 is one rate per second — audio, references and ratio add $0.
• 5s: $0.25 at 480p, $0.49 at 720p, $0.99 at 1080p
• Prime is about 1.5× — start with standard
• 30s at 1080p is $5.93, versus $21.10 on Seedance 2.5
Guide 👇 https://t.co/XX4oPlgqkV
AnyModelBot bots can hand work to each other.
• Say "hand this to Product Assistant"
• That bot uses its own model and budget
• Approvals stay in the same chat, under its name
• The first bot checks the result, then reports back
Open source (MIT) 👇
https://t.co/f5gfX3HAoP
A successful AI call can hide a failed first attempt.
Trace the request across every try:
• One logical request ID plus an attempt number
• Record route, model, timings and terminal state per attempt
• Don't stitch streamed fragments by array position
• Allowlist telemetry — no prompts, secrets or raw tool args
Guide 👇 https://t.co/j1bwIxWmEx
Flare and Sunburst share one price. Choosing between them isn't about money.
• Flare at medium is the everyday default
• 1536×1024 Flare runs settled at ~$0.011 (estimate $0.044)
• GPT Image 2 is ~29% cheaper per token, fewer controls
Guide 👇 https://t.co/HNq6ZnlhxD