AI agents
An AI agent you can trust with real work, in about a week.
An assistant that answers from your data and takes actions in your tools, with guardrails, logging, tests and a monthly cost ceiling. Not a demo that makes things up.
Prefer to write? Send your requirements
In short
PutThrough builds AI agents — customer support agents, lead qualification, document and inbox triage, and internal knowledge assistants — in about a week, at a fixed price agreed in writing after a free scoping call. Each agent does one job, answers from your own data with sources, calls only the tools you approve, hands off to a person when it isn’t sure, and runs under a monthly cost ceiling with logging and a test set it must pass.
01Who it’s for
For teams with repetitive questions and clear rules.
-
Support teams answering the same questions
An agent that answers from your help docs and order data, and hands off to a person when it isn’t sure.
-
Sales teams sorting inbound leads
Every inquiry qualified against your criteria, enriched, and routed to the right person with a summary.
-
Teams buried in documents or inboxes
Incoming email, forms or files read, labeled and routed, with the uncertain ones left for a person.
02Example uses
What a one-week agent usually is.
-
Customer support agents
Answers from your docs, looks up orders or accounts through tools, and escalates with the conversation attached.
-
Lead qualification
Asks the questions your sales team would, scores the answers, and books or routes the good ones.
-
Document and inbox triage
Reads what comes in, extracts what matters, and files or routes it, with a review queue for anything unclear.
-
Internal knowledge assistants
Answers your team’s questions from your own documents, with sources, and only from what each person is allowed to see.
03What fits in a week
One agent, one job, with guardrails.
Fits in a week
- Answers grounded in your documents, with sources
- Two or three tools it can call, such as order lookup
- Handoff to a person when it isn’t sure
- A monthly cost ceiling and usage logging
- A test set of real questions it must pass
- Chat on your site, or inside a tool you already use
Needs longer, or a call first
- Actions that move money without human approval
- Several agents coordinating with each other
- Training or fine-tuning your own model
- Formal compliance programs such as HIPAA or SOC 2
04A sample week
Seven days, one agent.
The same seven days as every build, with this product’s work in them. Your week is planned on the scoping call.
- Day 0
Scoping call and fixed quote
The one job the agent does, what it may and may not do, and the questions it has to get right.
- Day 1
Data, tools and test set
Which documents and systems it uses, the tools it can call, and a written set of test questions with the right answers.
- Days 2–5
Build, in the open
The agent, its tools and its guardrails, scored against the test set every evening, with a preview you can try.
- Day 6
Harden
Misuse and prompt-injection cases tested, the cost ceiling enforced, logging and alerts switched on.
- Day 7
Launch and handover
Live where your users are, then the handover: how to add knowledge, read the logs and change its rules.
05What you get
What you walk away with.
- The agent, live on your site or in your tools
- The test set, and the results it passed at launch
- A monthly cost ceiling and usage logging
- A handoff path to a person, with the conversation attached
- A guide to adding knowledge and changing its rules
Plus the standard handover every build ends with: source code, a written handover, a recorded walkthrough and every account in your name. The full Day 7 list
06Stack
Swappable models, boring infrastructure.
Models and prices move fast, so the model sits behind an interface you can swap. Postgres with pgvector before a dedicated vector database: one fewer system to run.
- Models
- Claude or OpenAI, behind one interface
- Language
- TypeScript
- Knowledge
- Postgres with pgvector
- Quality
- A test set, tracing and cost limits
07Pricing
A fixed price, and a fixed ceiling on running costs.
Your exact price is fixed in writing after the scoping call, before you commit to anything.
After launch, the care plan is from $499 per month.
Prefer to write? Send your requirements
08Questions
01 Is this the same as an AI chatbot?
A chatbot answers questions. An agent can also act, such as looking up an order, creating a ticket or booking a meeting. We build both, and recommend the simpler one when it does the job.
02 What stops it making things up?
It answers from your documents and shows its sources, it is tested against real questions before launch, and when it isn’t sure it hands off to a person instead of guessing.
03 What will it cost to run?
That depends on how much it is used, which is why every agent has a monthly cost ceiling you set. We estimate running costs on the scoping call.
04 Is our data used to train models?
We use providers’ business APIs, whose terms exclude your data from training by default, and we tell you which provider and settings are used.
05 Can it take actions, not just answer?
Yes, through tools we define together, such as looking up an order or creating a ticket. Anything that moves money or can’t be undone waits for a person to approve it.