We are not a chatbot platform.
We are the layer beneath.
This page is shorter than most. Here's why we exist, what we believe, and what we will never do — written without ceremony, because you don't have time to read it twice.
Three years ago, building an AI agent meant duct-taping six things together: a model from one vendor, a prompt library from another, a RAG store from a third, audit logs in a fourth, a human-in-the-loop hack in a fifth, and a brittle webhook holding it together. We watched smart teams ship agents that worked in demo and broke the moment a real customer asked something off-script. We watched compliance teams shut down promising pilots because nobody could replay what the agent did at 3:14 a.m. on a Tuesday.
The agentic part — the model — was never the bottleneck. The plumbing was. The plumbing is.
So we built the plumbing.
Thaliq is the layer between your product and the model. Multi-model routing because you should not be married to whoever wins next year. MCP gateway because your stack already exists and you should not have to rebuild it. Human-in-the-loop because some decisions belong to humans and pretending otherwise is malpractice. Audit by default because your future self deserves to understand what your past self shipped.
We are based in LATAM. The last decade of B2B SaaS was designed for the San Francisco bubble — pricing in USD, contracts in English, support windows aligned to PST. We watched founder after founder in México, Lima, Buenos Aires and São Paulo bend their roadmap to fit tools that were not built for them. Thaliq is the bet that LATAM should have its agentic layer without translation. Built here, priced here, supported here — usable everywhere.
We started small on purpose. We will scale only as fast as we can stay honest about what works and what does not.
Five things we hold non-negotiable.
- 01Lock-in is theft.
Switching costs that exceed three days of engineering are a tax on your roadmap. We design for portability before we design for stickiness.
- 02Audit is not optional, ever.
If you cannot replay the turn that triggered the refund, you do not own your agent. We log every model call, every tool result, every HITL decision — from day one, not from when compliance asks.
- 03The model is the easy part.
Anyone can call an LLM. The hard part is the policy around it, the permissions, the failover, the cost ceiling, the audit. We obsess over the boring middleware.
- 04A demo is not a product.
The agent that works in the founder's hands at the Friday demo is rarely the agent that survives Black Friday. We build for production first; demos come for free.
- 05Your customers belong to you.
We do not train on your conversations. We do not share your prompts. We do not aggregate your usage to sell back as "industry benchmarks." Your data goes in, your data comes out, and that is the entire contract.
What we will never do.
- ×Train models on your customer data
- ×Lock you to a single LLM provider
- ×Pretend our agents are human
- ×Hide pricing behind a sales call
- ×Ship features without an audit trail
- ×Bury bad news in a changelog footnote
- ×Promise "enterprise-grade" and mean "we have a Slack channel"
If any of these change, this page changes first.
Built in LATAM, for everyone.
We are a small team based in Lima, with operations across the region. We speak Spanish and English at the same level. Our hours are LATAM business hours. Our contracts can be in PEN, MXN, COP, ARS, BRL or USD.
Our customers are in México, Chile, Perú, Colombia, Argentina and Brasil — and increasingly in the US and Europe via the embedded SaaS route.
If your company is in LATAM and you are tired of explaining your context to a vendor that has never operated outside of San Francisco, we built this for you. If your company is anywhere else and you want a partner that thinks across timezones because it has to — same.
Talk to us.
Email is the best way. We read every message and reply within one business day, from a real person on the team.
No sales sequence. No nurture campaign. Just a thread.