It seems that most guides, not just on here, seem to prefer Telegram as the default interface to interact with your agent. I frankly hate Telegram. The few times I've used it, I was just inundated with spam and scammers, and I'm much more comfortable with Discord. Granted I've never run my own server on either until now, so maybe that would make Telegram bearable. That being said, I've found myself working primarily directly with my agent(s) via the WebUI chat or doing things myself via the terminal (either my server or the openclaw container). Now that I seemed to have tamed my main agent's persona thanks to help from some of the guides on here, I'm ready to start trying to implement some of the more advanced concepts (CTO Agent, Notes Agent, etc.), and I'm wondering if I should pivot to Telegram or just stick it out with Discord. I've already had my agent (Skippy) download and review the code of the CTO Agent, and it says migrating it to Discord "shouldn't be too difficult." So before I start banging my head off that tomorrow, I thought I would see what other people think.

Now that I've been tinkering with OC for a few weeks, and I (or my agent) have broken more things than I care to admit. How does everyone tend to makes changes to their agent(s)? Even the CTO Factory Agent, while much better then my original agent, still breaks stuff from time to time. I been mostly using the agents themselves to make config changes, but unless you're on a top tier model (and even then) it can be problematic. I also use the OC terminal for some things, but that tends to break some of the customizations I'm running. Then I also sometimes edit the .json files directly. Mainly to recover from a catastrophic failure (i.e. OC won't start). I haven't used the WebUI for editing, but that IS how I mainly interact with my agents. But aside from the chat, which isn't great, the rest if the WebUI is pretty clunky IMHO. So just looking to see how other people do it.

Voice in → Voice out. We're improving our Riff agent by teaching it to speak. Part 8 is live!

Heads up: Every member of our team has been hitting massive usage spikes in Claude Code today. In some cases, a full 5-hour limit was completely exhausted in just 15 minutes. This is likely due to documented caching bugs where the system fails to reuse context and instead re-bills for the entire codebase on every turn. We are still studying the issue, but the following steps have already helped us mitigate the drain: - Strict .claudecodeignore: Add everything non-essential (node_modules, dist, .git, logs). The smaller the context Claude "sees," the slower the drain. - Frequent /compact: Run this command every 10-15 minutes. It truncates session history and forces a fresh, smaller cache point. - Direct File Paths: Instead of asking Claude to "search" or "find" things, specify exact paths (e.g., /src/components/Button.tsx). This prevents the tool from burning tokens on repository exploration. - Restart for New Tasks: Exit and restart the CLI when switching focus to prevent "context bloat" from carrying over. Technical context for those affected: - GitHub Issue #40524 - Reddit Deep-Dive on Cache Bugs We are continuing to monitor the situation and will keep you posted as we find more ways to optimize usage. Stay tuned!

Step 3 is live 🔥 : Option 1: Aggregators as your token control panel This is the practical setup for OpenRouter: how to find strong :free options and lock your future model IDs for config. Drop your questions below and I’ll answer each one. 😊