The only AI router with observability & evals on every call. Creator of Span-02. OSS gateway: https://xitter.twitterapi.xyz/t.co/1Htxa4SA9u Community: https://xitter.twitterapi.xyz/t.co/Q2UX2AO8Zf
San Francisco, California
Joined March 2024
- Tweets850
- Following80
- Followers3.5K
- Likes1.6K
Pinned Tweet
Introducing Span-02: anytime reasoning for decisions.
Reasoning models make you pick an effort level and wait. Span-02 takes a deadline and always returns its best decision so far.
It gets better the longer it thinks. With reasoning off, it beats Jev at 1/4 of the price, and runs at 1/3 the latency of decision models at its level.
It can:
- Answer with numbers and scores, not just probabilities and choices
- See images and screens (multimodal)
- Take a configurable reasoning time
And it’s fully open weight!
Respan retweeted
Retell just crossed $100M in revenue run rate.
$10M to $100M in 17 months. Profitable. $4.6M raised.
From day one, we’ve focused on solving the hardest parts of deploying voice AI in the product itself, so customers can build, manage, and improve their own agents without depending on us.
Our bet: the future of the contact center belongs to the teams who run it.
Proud of our team. Grateful to every customer who believed in us.
Respan retweeted
We trained a custom model to kill the only moat CRMs have, switching cost.
@RespanAI asked @octolane to migrate them off Attio to Octolane. Octolane migrated years of their data in under 4 minutes.
That’s Exodus 1, our in-house migration model. It reads your old CRM’s schema, maps every custom field and object, carries over full conversation history, and enriches every account and contact in flight.
Attio customers now migrate in under 5 minutes. HubSpot customers in under 20.
And we’re already training Exodus 2 to migrate Salesforce customers in under 30 minutes.
Respan retweeted
we're making Gemini 4 Carbon free on @RespanAI
google's best model yet, and it's better than Opus 5.5
no card, no credits, just an API key
48 hours only. try it now
platform.respan.ai/platform/…
Respan retweeted
hi @X i'm Kelly! sf-based, looking for more friends in tech :)
a lil about me:
> journalism grad
> used to make documentaries
> now making launch videos at an AI startup
> learning to code with claude code, slowly
would love to meet anyone building cool things, especially if you still shoot film for fun 📷
Respan retweeted
Prompt caching: why does Anthropic only offer 5-minute and 1-hour options?
It comes down to where your cache is stored.
A cache can sit in GPU memory (fast, expensive) or CPU memory (slower, much cheaper).
Either way, keeping it costs money every minute. At some point it costs more than just rereading your prompt.
I did the math. The break-even point:
~5 minutes on GPU (HBM)
~1.5 hours in CPU memory (DRAM)
So when you pick 5 minutes or 1 hour, you’re really telling Anthropic where to store your cache.
Respan retweeted
we’re making Muse Ember free!
Meta’s newest and most cost-efficient model built for high-volume agent work.
no payment or credit card required.
1. create an API key on @RespanAI
2. call meta/muse-ember:free
fyi: free usage has a token limit. deposit $5 to raise it (still free).
Today we're introducing Muse Ember, our smallest and most efficient model, built for complex reasoning execution and everyday agent work.
Muse Ember can carry workflows across apps, call tools in sequence, run tests and make targeted code fixes. With a 1M-token context window and native understanding of text, images, audio and video, it can also work across large document sets and varied inputs.
Weekly drop #4 is live on the Respan AI Gateway.
Top deal: Hy4, 70% off. $0.25 in / $0.75 out per 1M tokens.
Also this week:
GPT-6 Astra: 18% off
Claude Sonnet 5.5: 15% off
Qwen 3.8 Flash, Qwen 3.8 Max, Kimi K3.1 Pro: 30% off
MiniMax M3.1 Flash: 20% off
Gemini 3.8 Flash, Gemini 4 Argon: 50% off
DeepSeek V4.1 Flash 1017: 30% off
No transaction fees. Full prices in the comments.
New drop every Monday 🧵
Respan retweeted
shipped a small one today: redesigned the Models page on @RespanAI
recommended and discounted models are now grouped together and pinned at the top, so you can spot them at a glance
also open to any feedback on the gateway. if we adopt your idea, i'll send you $100 in credits
Respan retweeted
Respan Gateway runs on my Raspberry Pi 4: 20 MB of RAM, about 2 ms added to first-token latency at 100 concurrent streams.
At 256 streams, the 2 cores I gave it were maxed out: still only 35 ms added on a 1.8 GHz CPU.
I ran a 15-minute stress test: 40,718 requests with zero failures.
Now I kinda want to get a few more Pis for load balancing 😂
Respan OSS Gateway is also an MCP gateway.
Add your MCP servers to one config and every client connects to one URL.
- Credentials stay on the gateway
- Tools switch off without restarting clients
- Every MCP request can show up in Respan.
A reverse proxy isn’t enough. Each server needs its own session, and two servers can both have a search tool.
So the gateway holds a session per server and prefixes tool names: github__search, linear__search.
Tencent Hy4 is out, and Respan is the cheapest place to run it.
60% off list:
→ $0.3336/M input
→ $0.0168/M cached input
→ $1.0004/M output
That's 20% below every other gateway - and no platform fee on top.
Live within 30 min of release.
respan.ai/models/hy4
Respan retweeted
I used to film documentaries waist-deep in the ocean. now i film ai startup launch videos in a garage
documentary vs launch video:
> 130 hours of footage → 45s of screen recording
> waiting for golden hour → waiting for deploy
> b-roll of the ocean → b-roll of a cursor
> people who are shy on camera → engineers who are shyer
> blurring faces → blurring API keys
> the hero is a person → the hero is a dashboard
the job never changed: find the 10 seconds that make a stranger care
Respan retweeted
Respan Gateway (OSS) does server-side tool calling.
no tools in your client. no tool loop to write. the gateway calls them and feeds the results back to the model.
also works in realtime voice, so the model can use a tool mid-conversation, even over OpenAI Realtime.
Weekly drop #3 is live on the Respan AI Gateway.
Top deal: GPT-6.1 Terra, 30% off. $0.70 in / $3.50 out per 1M tokens.
Also this week:
Claude Opus 5.5: 7% off
DeepSeek V4.1 Flash 1017: 55% off
GLM 5.3 Flash: 50% off
Hy4 Preview: 40% off
Kimi K3.1 Pro: 30% off
Claude Haiku 5.5: 15% off
GPT-6.1 Sol: 15% off
GPT-6 Luna: 15% off
0 transaction fees. 0 markup.
New drop every Monday. Full prices in the comments.
Respan retweeted
10,000+ lbs of cookies just hit Costco shelves across the bay area 🍪
the logistics startup behind the deliveries runs its AI features on the @RespanAI gateway.
we spend our days staring at tokens and latency charts. seeing them turn into cookies you can actually buy hits different.