I’ve been heads-down for the last few weeks. Here’s what’s new.
See Your Agent Thinking
When your agent is working through a response, calling tools, looking things up, pulling data, you can now watch it happen in real time. The chat widget shows a live activity feed so there’s no more staring at a spinner wondering if something broke.
Scheduled Tasks
Set a schedule, write a prompt, and your agent runs it automatically. Think daily reports, weekly summaries, inventory checks. Results get delivered to Slack or email.
“Send me a summary of support tickets from the last 24 hours, grouped by priority.” Every morning at 8am. Done.
Ask Questions About Your Data
This is a big one. Your agent can now query your datasets using plain English. No SQL, no filters, no exports. Just ask.
“What products do you have between $300 and $500, that are in stock and green?”
Behind the scenes the agent translates your question into a structured query, runs it against your indexed data, and comes back with an answer. It can discover what fields exist, what values are available, and paginate through large result sets. It’s like giving everyone on your team a data analyst who never sleeps.
HTTP & S3 Document Sources
Documents and datasets can now be pulled from HTTP endpoints and S3 buckets. Point your agent at a URL or a bucket path and it’ll fetch, index, and keep it up to date. Handy for product feeds, knowledge bases, or any data that lives outside your app.
File upload still works of course, but when integrating with external systems it’s often just easier to pull sources from a bucket or server. S3 and S3-compatible services should work* (tested on Cloudflare R2).
MCP OAuth 2.1
Connecting to remote MCP servers just got easier. Full OAuth 2.1 support means your agents can securely authenticate with third-party tools without you having to manage tokens manually. Some vendors periodically require re-auth, so I’ve added that from the workspace credentials as well.
Smarter Slack & Email
Slack replies that were too long used to get cut off. Now they split into multiple messages automatically. Tables render properly. Duplicate events are handled gracefully. And you can now set a channel-specific prompt so your agent behaves differently in Slack than it does on your website.
GPT-5.4 Support
GPT-5.4 is now available as a model option. In my testing, I’ve found Claude Opus 4.6 and Sonnet 4.6 handle tool use a little better than GPT-5.4, but your mileage may vary. How’s your experience stacking up? Drop me an email to let me know.
Reliability Under the Hood
Agents are now more resilient when things go wrong. If a provider has an outage, the system backs off and retries intelligently instead of hammering the API. Permanent errors are caught early so your agent doesn’t burn through credits on something that’s never going to work.
A Cleaner Experience
The onboarding flow is smoother, the personality page brings together everything about how your agent sounds and thinks, and the activity dashboard now shows real usage data. Actual conversations, actual users, actual token spend. I’ve also tightened up security across the board.
More coming soon. As always, if you have questions or feedback, I’d love to hear from you.