Blog
Thoughts on AI agents, MCP infrastructure, and building services for the agent economy.
Slates, Bumpers, End Cards, and Joining Clips: Conform Learns to Splice
A still image on the front of a video, a title card in the middle, or two clips joined into one — three different names for the same underlying operation. We added conform_splice_from_url and conform_splice_media: an ordered list of video and/or image segments in, one continuous fresh-encoded output out.
The Missing Step Between 'AI Generated the Video' and 'Here's the Deliverable': Introducing Conform
Generative video pipelines routinely hand you a silent picture track and a separately-produced audio track — a voiceover, a dub, a music bed rendered by a different tool entirely. Muxing them back together sounds trivial until the frame rates don't match or one track runs three seconds long. We added conform_from_url and conform_media so an agent can do this in one call instead of hand-rolling ffmpeg.
Three People, Eight Speaker Labels: What Multi-Speaker Calls Do to Transcription
We ran a 57-minute three-person interview through a major meeting platform's built-in transcription and through Botverse Transcribe. The built-in version produced eight speaker labels for three people, invented a participant who was never on the call, and dropped the three words that carried the entire point. Plus: why you can't just paste the audio into a chatbot, and what the transcript costs you in tokens if you try.
From a 60-Minute Recording to a Folder of Highlight Clips, in One Job
Transcribe a video, and whenever someone says a keyword, cut a 30-second clip there and drop the clips plus the transcript in a folder. It sounds like a script you'd babysit for an afternoon. With Botverse it's two tool calls and a poll loop — and it shows exactly where an agent's job ends and a workflow engine's begins.
Botverse Transcribe: One Call, a Named Transcript, Done
Speech-to-text is a commodity. The hard part is knowing who said what and getting a usable document out the other end. Botverse Transcribe is one call: point it at a video or audio file, hand it the attendee names, and get back a clean, speaker-labelled transcript — txt, json, captions, or a formatted Word/PDF. We tested it against a commercial notetaker on a real 60-minute, 17-person call. It named 10 of 11 speakers correctly.
Introducing Botverse: Infrastructure for the Agent Economy
AI agents are becoming capable of real work. What they still lack is the infrastructure to do it efficiently. Botverse provides core processing services any agent can call — without needing custom deployments, no applications, no S3 setup, no infrastructure to manage.
Agents Shouldn't Be Your Compute Layer
Every token your agent spends on file I/O is a token it isn't spending on the problem you hired it to solve. Offloading transactional work to specialist services isn't just faster — it makes agents smarter.
Why We Built Botverse: The NearlyMe Story
NearlyMe is an AI executive assistant we built internally. As it grew, it needed to handle real media workflows and document processing. We learned a lot about agentic performance and it drove the need to build Botverse.cloud.
The Document Context Tax: What You're Really Paying When Your Agent Reads Files
When an agent reads a document inline, all those tokens sit in the context window for every message that follows. For a 3,500-word report, a 10-message session burns 52,000 tokens just from the document sitting in memory. We did the numbers — and they're worse than you think.
One Submit Call, Five Outputs: How Botverse Workflows Replace Multi-Step Agent Pipelines
A conference recording needs to become an MP3, a timestamped transcript, structured meeting notes, an action items spreadsheet, and a thumbnail. In a typical agent pipeline that's five sequential steps with the transcript in context the whole time. In Botverse it's one workflow definition, submitted once.