Building @Brevitas_sys | Saving you costs on inference workflows | Prev. App Dev @Apple | @fdotinc Off Season II Fellow | @join_ef S26 Fellow
San Francisco, CA
Joined December 2022
- Tweets229
- Following138
- Followers143
- Likes645
Pinned Tweet
Today, we're introducing Splice.
AI agents are constantly rebuilding context from scratch, even when most of that context hasn’t changed.
So we built a way to edit the model's KV cache directly. Add context, remove it, or change it. Keep the work that’s already been done.
Splice is our first step toward making model memory programmable.
Built with ❤️ by @Brevitas_sys
Check it out: brevitassystems.com/blog/spl…
Anish Paleja retweeted
We raised $1.5M from Y Combinator & Airtree Ventures to transform debt capital markets.
Good deals get stuck in the wrong inboxes.
Advisors can’t know every lender. Funds can’t read every deal. And what a lender wanted last month may not be what it wants today.
Forward connects private credit funds with the leading debt advisors. – at San Francisco, CA
Anish Paleja retweeted
The cloud providers killed our machines 69 times, but we didn't lose a single job.
Our coding agent used Portal to run all 54 jobs in our demand forecasting project on spot instances, cutting our cloud bill by 3x ($124 → $43).
recomputing an entire LLM context because a few tokens changed feels incredibly wasteful.
today we benchmarked Splice on an A100 and hit up to 13.57x faster edits on Qwen2.5-7B in fp32, with matching tested continuations.
back to building.
Anish Paleja retweeted
We rented the same H100 for 80% less.
The catch: the cloud can take it back at any moment, and your job dies with it.
So we built Portal.
It saves your running Python job and resumes it on any other machine, any cloud, mid-run. No code changes. One-prompt install and forget.
Anish Paleja retweeted
Today we’re launching WonderDrive by @wondersearchai (YC F26): one drive for every agent.
Close your laptop and Claude Code picks up where you left off. Codex picks up where Claude Code left off.
Your agents run in the cloud, and your files live on your Mac. WonderDrive makes them one drive, so you and every agent work on the same files.
→ Hand off between you, Claude Code, Codex and Cursor without copying a single file
→ Every agent works on the same files at the same time
→ Built-in search, powered by WonderSearch
→ Run and deploy straight from the drive
→ Rewind any change when an agent does an oopsy
Coming soon to every WonderSearch user.
Your agents just got a home.
@ycombinator @fdotinc @daltonmeon
Anish Paleja retweeted
We just gave Claude Code, Codex and Cursor one shared brain.
One learns something, they all know it. One deletes 48,120 files, you hit undo.
No vector database. No embeddings. We matched OpenAI’s best embedding model without using a single one.
Introducing WonderDrive: one drive for every agent you run. Coming soon. wondersearch.com
@ycombinator @fdotinc @daltonmeon
Anish Paleja retweeted
I recreated subvocal in 9 minutes with one prompt for AirPods Pro 3.
Source: github.com/joeblau/subvox
We’ve raised $1.3M from YC & Afore Capital so you never have to type or speak again.
Typing is slow. Talking into your computer is awkward. Subvocal fixes both
Our glasses give you the speed of speech without you making a sound or touching a keyboard.
It tracks your internal speech muscles with non-contact sensing and decodes silent, almost invisible mouthing into text at more than 200 words/min.
Pre-orders are now live at:
subvocal.co/preorder?utm_sou…
Anish Paleja retweeted
We’re about to make the entire RAG industry obsolete.
We matched OpenAI’s best embedding model without a single embedding. Now we’re replacing the whole stack with one drive.
Today, we're announcing WonderDrive. One drive every AI can read. Claude, GPT, Gemini and your own agents all work from the same files, so one picks up exactly where another left off.
Coming soon from @wondersearchai
@ycombinator @fdotinc @daltonmeon
Anish Paleja retweeted
@daltonmeon and @daleverett went from building in a one-star motel in SF to @ycombinator's F26 batch to build @wondersearchai
Their rule? Get the heck off localhost.
Anish Paleja retweeted
A lot of you have been asking how Splice works.
So we made a quick breakdown
A lot of you were wondering how Splice actually works.
Here's a quick 60 second explainer
We've been building something pretty crazy with Splice.
It lets you edit an AI model's memory directly, deleting, inserting, or moving context without recomputing everything from scratch.
Made a quick explainer of what it actually does
3
268
Anish Paleja retweeted
A quick explainer on what we’ve been building with Splice.
The idea is pretty simple once you see it.
A quick explainer on what we’ve been building with Splice.
The idea is pretty simple once you see it.
Anish Paleja retweeted
We've been building something pretty crazy with Splice.
It lets you edit an AI model's memory directly, deleting, inserting, or moving context without recomputing everything from scratch.
Made a quick explainer of what it actually does
These guys are moving faster than the speed of light
What happens when your data changes faster than your search can keep up?
WonderSearch 1.1.
Search your data as fast as it changes.
No embeddings required
+7.9% average benchmark gains
Coming soon to @polygres (YC F26)
link in first reply ↓
3
145