Agents that work while nobody is there, and stop where you decided.
Your rules live in the Corpus, the agent runs on Line, and every run is kept in Semant. Start with any of the three, or ask us to build it for you.
Saturday, 22:14. Nobody is in the office.Saturday, 22:50. Nobody is in the office.A recorded night of the sample agent
Recorded runThe agent asked the Corpus
"damaged part replacement policy"
and the Corpus returned the written rule
Replacementsdoc-replacements · version 4
A part that arrives broken within 12 months of delivery is replaced. A person confirms the part against the order record before it is shipped; the customer hears back within 5 business days.
What the agent may never do
- ship a replacement without a person confirming the part
- share another customer's order details
A customer wrote22:14:0322:50:03
The agent drafted
We're sorry a part in your order #48213 arrived broken. We'll ship a replacement of part #A-2291 within 5 business days [1].
It did not send it: the draft promised a shipment before a person confirmed the part. It stopped and asked the person on shift.
You are on shift. What do you answer?
Approved at 22:20:15. The exact draft you read went out, same hash.
4742e901bc3bc30eb92ba379…
Denied at 22:56:15. The draft did not go out, and the customer got this notice:
A person from our team will follow up on your replacement within one business day.
What Semant kept, step by step
- turnmessage
- triage1 model call
- retrievecorpus.search, corpus.get_document ×2in the Corpus
- draft1 model call
- check1 model call
- draft1 model call
- check1 model call
- escalateescalate
- Waiting for an answerWaited for a person until 22:20:15Waited for a person until 22:56:15
- resume_agentanswer: approved
- escalateescalate
- replyreply
- close
- resume_agentanswer: denied
- escalateescalate
- close
trace 38b382e569b438ab… · 24 spanstrace 2b1c164e424eb7a3… · 22 spans
This is Line's code with a test model, recorded for this page. Each answer has its own recorded run.
Three products. Start with the one you are missing.
Each one works on its own and talks to the others through its public API.
Your documents and your rules, served to your agents. The same question gets the same answer, with the rule that decided it.
Start with the CorpusPOST https://corpus.trysenda.xyz/v1/experts/deal-desk/query
Same question, same traceHash. Copy the curl from the Corpus page and compare it from your own terminal.
You build the agent in the browser, without code: what it may touch, what it may never do, and whom it asks. You stop it whenever you want.
Join the waitlistCases of the sample agent, version 3
Answered with a source 3/3 passes Cited the expected document 2/3 fails Stopped when it had to 5/5 passes Used nothing it may not 5/5 passes Left a complete trace 5/5 passes Does not launch: one criterion fails.
A version launches only when every case passes. This one did not.
What each agent did while you were not looking, step by step, even if it was not built on Senda.
Join the waitlistOne night, 22:00 to 06:00
Two recorded runs and five replayed from them. The gap is on the record too.
One request, several agents, and a person who decides.
Watch how they share the work: they read, draft, check, and stop to ask when it matters. Everything is written down.
A request arrives
A customer writes in and the request enters the team of agents.
They split the work
One receives the request and divides it: an agent reads the documents and another drafts the answer.
They pass along what they found
The reader hands over what your rules say. The Corpus decides with them and gives the same answer every time.
They check each other
Another agent controls what was drafted before it goes out.
They stop and ask a person
When judgment is needed, Line holds the work and waits. Nothing goes out until someone answers.
It continues and is written down
With the answer the work goes on, and Semant keeps every step so you can review it later.
No time to build it? We build it for you.
Tell us which work you want to hand over and we build it on the same products you see here. We answer with the scope and the price.
- It runs on Line and reads your rules from the Corpus.
- Every run is kept in Semant, where you can see it.
- The access is yours: you grant it, you revoke it, and you stop the agent whenever you want.
Write to ushello@trysenda.xyz
Before you start.
Is the run above real?
It is a real run of the sample agent: Line's code, with a test model in place of the real one, recorded for this page. It is not live, and each answer has its own recorded run.
How much does it cost?
Each product has three tiers: Basic and Pro, which you subscribe to yourself, and Enterprise, agreed with us. Line adds credits for what your agents consume. There is no free plan. The Corpus prices are on its page.
Do I need all three?
No. The Corpus works on its own, Semant records agents built with any tool, and Semant comes with Line.
Do I need to write code?
Not for the rules or the agent: documents are written and the agent is built in the browser. Someone technical connects the API when you want to put it in front of your customers.
What does the agent do when it does not know?
It says so instead of guessing. In Line it stops and asks the person on shift, who approves or denies the exact draft that would go out.
Where does my data live?
On Senda, each product in its own database, and you can read your documents through the API whenever you want. Semant keeps steps and tokens, never prompts, answers or documents.
Choose where to start.
The Corpus can be bought today. Line and Semant have a waitlist. And if you would rather not build it yourself, we build it for you.