01
Instructions that travel
Set your voice, your stack, your constraints — once. They prepend every chat. T1 still picks the model; your instructions follow.
Meet Oran T1 · The routing model
No single AI is best at everything.Oran T1 is.
T1 routes each message to the model best-suited to answer it. Reasoning, code, writing, vision — every question gets the right brain, automatically.
· Inside Oran T1 — routed per question ·
Anthropic
- Claude Sonnet 4.6
- Haiku 4.5
OpenAI
- GPT-5.4 mini
- GPT-4.1
- o4-mini
Google
- Gemini 2.5 Pro
- Gemini Flash
Why routing
Different models are best at different things.
T1 is best at picking which.
Claude
GPT-5
Gemini
o-series
Haiku
T1 picks
Reasoning
→ o-series
Writing
→ Claude
Code
→ GPT-5
Long context
→ Gemini
Speed / cost
→ Haiku
Oran T1routed max
every row
Illustrative — bars reflect typical relative strengths from public evaluations (HumanEval, MMLU, MT-Bench, long-context retrieval). Detailed sourcing in our routing methodology.
Live in oran.chat
One chat. Three brains, mid-thread.
Why does my Postgres query take 3 minutes when EXPLAIN says 200ms?
T1 → o4-mini · reasoning
EXPLAIN gives the planner's estimate. The 200ms cost is the predicted path. Three minutes in practice usually means the actual rows returned are 100–1000× more than estimated. Run EXPLAIN ANALYZE to see real row counts — likely a stale ANALYZE on the join table…
Great. Now write a polite Slack to the DBA asking them to re-run ANALYZE on the join table.
T1 → claude sonnet 4.6 · writing
Hey Sam — quick favor. The orders × line_items query is hitting 3+ min and I'm pretty sure ANALYZE is stale on line_items. EXPLAIN ANALYZE shows real rows ~400× the planner estimate. Could you re-run ANALYZE when you get a sec? No rush. Thanks for the help!
query-plan.png
Quick — what's this output mean?
T1 → gemini · vision
It's a nested loop join over 412K rows on the outer side. The inner index scan is fine, but the loop count is the killer — every outer row triggers another lookup. A hash join would be ~50× faster here…
The room around the model
Everything T1 needs to stay useful,
turn after turn.
02
Branch instead of overwrite
Don't like an answer? Regenerate creates a branch off the same parent. T1 picks again — often picks differently — and both paths stay alive.
03
Bring the source material
Drop in PDFs and screenshots. When you attach an image, T1 routes to a vision-capable model. Document context stays in the thread.
04
Web search with citations
On for everyone. T1 pulls fresh facts and cites them inline, instead of guessing.
Pricing
Start free. Pay for the volume you need.
Every message is automatically routed to the best model — you never pick. Free is capped daily; paid plans lift that cap for a monthly allowance.
Free
Enough to get a feel for it.
$0forever
10 messages every day
- Every message routed to the best model
- Web search with citations
- Custom instructions, branching & files
- No credit card
Lite
Casual use, no daily ceiling.
$5/ month
500 messages a month
- Daily limit lifted
- Every message routed to the best model
- Web search, files & custom instructions
Pro
For everyday, all-day use.
$20/ month
2,500 messages a month
- Everything in Lite
- Plenty of headroom for daily, all-day use
- Longer threads kept in active history
Power
Heavy, sustained workloads.
$100/ month
12,000 messages a month
- Everything in Pro
- Built for heavy, sustained workloads
- Priority during peak load
Max
Effectively unlimited.
$200/ month
Unlimited messages a month
- Everything in Power
- Effectively unlimited (fair use)
- Highest priority — for teams & power users
FAQ
A few fair questions.
How does T1 decide which model to use?
T1 reads the prompt — content type, complexity, context length, whether an image is attached — and picks the model best-suited from its allowed pool. The decision happens before the message is sent to any model. Every reply shows you which model T1 picked.
What if T1 picks the wrong model?
Regenerate. T1 will re-evaluate and usually pick differently the second time — and the original answer stays as a branch. If a category of question routes badly, tell us — that's a model the classifier can learn.
Can I force a specific model?
No. T1 owns the pick. If your workflow depends on choosing a specific model every time, oran.chat may not be the right shape — ChatGPT and Claude.ai both expose model menus directly.
Can I set custom instructions?
Yes. Set a system prompt in settings and it prepends every chat regardless of which model T1 picks. Your voice, stack, and constraints — preserved across routes.
How does branching work?
Regenerate a reply and the new answer becomes a branch off the same parent. The original stays available, and each branch can continue from there.
How does the free tier work?
10 messages every day. T1 routes within the lightweight pool — Haiku 4.5, GPT-5.4 mini, and Gemini Flash. No credit card and no trial cliff. Quota resets at midnight UTC.
Do you train on my conversations?
No. Conversations stay private to your account. The underlying labs do not train on API traffic by default, and we do not use your content for model training.
Still curious? hello@oran.chat
Begin
Type. T1 picks. Read.
No menus. No version numbers. No second-best answers.
Start free
No card. 10 free messages a day, forever.