Rish (@rishflips) header
Rish (@rishflips) profile picture

Sharing what works and what breaks when you build with AI agents. Building Twring: your agents can call you back. Free in beta https://teapawt.org/t.co/MoEP9PcSdm

Joined August 2011
Photo by Rish (@rishflips)Photo by Rish (@rishflips)Photo by Rish (@rishflips)Photo by Rish (@rishflips)Photo by Rish (@rishflips)Photo by Rish (@rishflips)Photo by Rish (@rishflips)Photo by Rish (@rishflips)Photo by Rish (@rishflips)
📞 Twring: call your agent, and let it call you back. I built Twring because I was tired of checking terminals and chat threads to see if my agent was done. Now my agent has a voice. Here's everything it does 👇 🗣️ Call your agents Press Call and talk to Hermes, Claude Code, Codex or Gemini CLI. It also works with any agent you can reach at its own web address. Your agent keeps its own brain, memory and tools. Twring just gives it a voice. 🔔 Your agent calls you When a job is done or it needs your OK, your phone rings. It rings full screen, even when the phone is locked, and you hear why it called. ✅ Approve by voice "Allow Hermes to push to production?" Say yes or no, or tap. 📝 Nothing said on a call gets lost Your agent writes your request down while you're talking. After the call, it works through the list and reports back in your chat. 🙅 Decline with a reply Pick Busy, Outside or Let's text, or type a short note. Quiet hours mean no agent can ring you while you sleep. 🤖 Works with AI apps too Add Twring as a custom connector in Claude, ChatGPT, Grok or Cursor. Then your AI can ring you or message you in Twring and read your reply. 🔗 twring.com/connector ⚕️ In the @NousResearch Hermes plugin marketplace Add Twring to Hermes like you'd add Telegram. Calls, voice notes, typed chat and results all stay in one thread. hermes plugins install twring 🏠 Self-host it One command puts Twring on your own Linux server, next to your agents. Your chats, files and agents stay on your server. curl -fsSL twring.com/install.sh | sh 🔓 Open source github.com/saasagelabs/twrin… 🔒 Private by default No recordings. Twring never runs your agent's tools. Keys are stored encrypted. Delete your data any time. 📱 iPhone and Android. Free while we launch. Ask for an invite 👉 twring.com Your agents are waiting for your call. 📞
14
10
708
Replying to @konrad_gnat
Try twring it lets you to talk with your agents you already run
2
Replying to @davidavenueai
Chats will improve in time soon, working on it!
1
1
8
Replying to @davidavenueai
Tnx bro, Tell me where it feels slow, confusing or broken.
1
1
13
Stop typing at your AI agent. Call it Talk the idea through out loud Say yes to the next step Hang up and get on with your day It keeps working. When it needs you, it rings Hermes agent, Claude Code, Codex, opencode... Free in beta ⬇️
1
1
5
168
Replying to @TheSimonHarms
Subagents per task would help a lot, one stuck task wouldn't sink the rest. Would you have a main agent check on them, or let each run on its own?
2
Caps in code, not in a prompt, is the right call. Prompts bend, code doesn't. What happens when a price changes a little between the yes and checkout?
1
1
29
400GB buys you a lot of runway. I'd still watch whether recall gets slower as the DB grows. Have you seen any slowdown so far?
8
Annoying when a plain ask doesn't stick. Putting the rule in its system prompt or a skill file might hold better than asking in chat. Worth a try?
7
Replying to @surlymochan
Mid-lie is a rough way to find out. What did it claim, and what tipped you off?
7
Replying to @epure_liviu
Per action, not per session, is the part that sticks. One yes covering ten later steps is just a key the agent holds. How does CounterSign keep the friction bearable?
3
Replying to @Hartdrawss
Slack summaries make sense for a team that big. When something truly can't wait, does Viktor have a way to grab you fast, or does it sit until you check?
5
Replying to @konrad_gnat
Hearing the briefing in your own voice is a fun touch. Do the bots ever talk back, or is it one way for now?
1
11
Replying to @guifav
4GB of cron noise in the session DB is a lot. Do you keep cron runs out of memory now, or still clean up after?
1
16
Replying to @epure_liviu
"Whoever can read the channel approved" is a sharp way to put it. What does the paper say agents should check instead?
1
13
Replying to @_anuragattri
A backup plan for usage is smart. When it hits a choice only you can make at 3 AM, does it wait or guess?
4
Replying to @_aashish_singh_
You can connect twring to your agents and let it ring you when it really need you
1
15
Blocking a swapped recipient after the yes is a really good catch. Is there a spending cap on each approval too?
1
2
81