AI agents that quote a price and stand by it.
React Web Development Services builds AI sales agents and support chatbots for Philippine businesses from ₱185,000 over 4–6 weeks. Messenger and WhatsApp commerce agents start at ₱95,000 (2–4 weeks), retrieval agents over your own documents at ₱280,000 (6–10 weeks), and internal operations automation at ₱140,000 (3–5 weeks). Running costs are the model provider's usage fees plus a care retainer from ₱9,500/mo. Based in Porac, Pampanga; we work across the Philippines.
The model talks. Code does the arithmetic.
Every AI agent that has embarrassed a business did the same thing: it was asked to generate a fact it should have been made to look up. A language model produces the most plausible next token, and a plausible price is not a real one.
So we split the job. Anything commercially binding — a price, a date, stock, an availability window — is computed by deterministic code and handed to the model as a value it is not permitted to alter. The model handles conversation, tone and clarifying questions. It never does the sums.
The Advisor on this site is built exactly that way, and you can test the claim rather than take it: ask it the same thing three times and the quote is identical, because a scoring engine produced it.
Ask a well-built agent the same commercial question three times and you get the same binding answer three times. If it doesn’t, it is not ready to talk to your customers — and no amount of prompt wording fixes it.
Sales & support agent
Qualifies, scopes, answers and books. Conversation design, CRM integration, escalation to a human, and a guardrail suite with a refusal path for anything it must not answer.
Messenger & WhatsApp commerce
Sells in the inbox where Filipino customers already prefer to talk. Usually the right first build if most of your enquiries already arrive on Facebook.
Retrieval over your documents
Your contracts, manuals, SOPs and policies made answerable with citations back to the source paragraph. Built on Claude or the model you prefer.
Internal operations automation
Quotes, approvals, onboarding and reporting. Often the higher-return build, because it removes admin nobody should be doing by hand.
Three cases where we will tell you not to.
Your volume is too low
Under roughly 30 enquiries a week, a shared inbox and a written reply template will outperform a ₱185,000 agent and cost nothing. The agent pays for itself on volume, and below that line the maths does not work.
Your answers aren’t written down
An agent can only be as consistent as the policy behind it. If your prices, lead times and refund rules live in three people’s heads and disagree, the agent will surface that disagreement at scale. Fix the policy first — that part costs nothing.
You want it to replace the team
It will not. It handles the repetitive 60–70% and escalates the rest, which makes a small team feel larger. If the business case only works by removing headcount, the case does not work.
Asked before every build.
How much does an AI chatbot cost in the Philippines?
An AI sales or support agent from React Web Development Services starts at ₱185,000 over 4–6 weeks, including conversation design, CRM integration and the guardrail set. A Messenger or WhatsApp commerce agent starts at ₱95,000 over 2–4 weeks. Retrieval over your own documents starts at ₱280,000 over 6–10 weeks. Those are published project prices, not estimates.
What does it cost to run each month?
Two things: the model provider’s usage fees, billed on your own account so you see them directly and we never mark them up, and a care retainer from ₱9,500/mo for monitoring, prompt and guardrail updates, and the conversation review that catches drift. Usage for a typical support agent runs a few thousand pesos a month at moderate volume.
Can you stop it inventing prices?
Not by prompt wording alone, which is why we do not rely on it. Every commercially binding value is computed in code and passed to the model as fixed. The model cannot alter it, so the failure mode is the agent saying “let me get a human” rather than quoting a number nobody agreed to.
Which model do you use?
Usually Claude, sometimes GPT, occasionally an open model where data residency requires it. It is a build decision, not a religion, and the architecture is deliberately model-agnostic so you are not stranded when pricing or capability changes. You own the prompts and the code either way.
Where does our data go?
To the model provider under a no-training agreement, and nowhere else. We sign a Data Processing Agreement under the Philippine Data Privacy Act (RA 10173), and if a class of data must never leave your infrastructure we design the retrieval layer so it does not.
Who owns it when we’re done?
You do — code, prompts, conversation designs, evaluation sets and the model accounts. Transferred on final payment, with no licence back to us and no dependency on us continuing. If you want to take it in-house, we hand it over and document it.