We build conversational chatbots with the Claude SDK connected to YOUR stack — Holded, Shopify, HubSpot, WhatsApp, Slack — with no black box and no dependency on opaque SaaS. Customer service, lead qualification, internal assistant. Your brand, your infra, your data.
Chatbot SaaS solutions (Intercom, Drift, Zendesk Chat) are powerful but have three problems that in 2026 are no longer unavoidable.
A pro plan on Intercom or Drift starts at hundreds of euros a month and scales with contacts/conversations. A custom bot with Claude has token cost (cents per conversation with prompt caching) + infra (tens of euros a month).
Chatbot SaaS come with predefined integrations. If your CRM isn't there, tough luck. A custom bot connects to ANY endpoint of yours via API: Holded, in-house systems, internal scripts. No limits.
The conversations stay in your database, not on a third party's servers. Useful for regulated sectors (accounting firms, clinics) and for having real control over GDPR and compliance.
These are the patterns we get asked for most. If yours doesn't fit, we assess it in the assessment call — the engine is the same, only the domain changes.
The chatbot converses, figures out whether there's a fit (budget, timeline, sector), collects contact details and opens an opportunity in your CRM with a structured summary. It doesn't promise timelines or prices, it gathers them for the human.
The customer asks 'where's my order?' over WhatsApp. The chatbot reads Holded, checks the shipment in Sendcloud/Correos, and replies with tracking and ETA. For returns, it collects the request and opens a ticket in Holded.
Your team stops having to open Holded for quick lookups. Ask in Slack: 'top-billing client this month', 'outstanding receivables for X', 'expenses in category Y'. The bot queries and answers with real data in seconds.
The bot answers 70% of frequent questions (tax forms, deadlines, typical deductions) by consulting your private documentation (RAG). When it doesn't know, it escalates to the human advisor with a summarized context. It cuts operational load without losing quality.
The bot knows your product (internal docs via RAG) and queries your API for user data (plan, usage, billing). It answers 60% of support tickets in seconds. The complex ones it opens as a ticket with enriched context.
It's not magic, it's engineering. These are the components we take care of in every chatbot we build.
Sonnet or Opus models depending on complexity. Native tool use to call your API, aggressive prompt caching to cut costs by 80%+ on repeated usage.
Your knowledge base (internal docs, FAQs, manuals) indexed with embeddings. The bot consults it before answering, with source citations. No hallucinating.
Holded, Shopify, WooCommerce, HubSpot, Pipedrive, Stripe, Sendcloud, WhatsApp Business, Slack — the whole catalog of integrations we already build.
The bot does NOT promise timelines or prices without human confirmation. It does NOT hit sensitive endpoints without auth. Full logging of every conversation for auditing.
Web embed (our own widget), WhatsApp Business API, Slack, Microsoft Teams, Telegram. Same bot, different surfaces.
On your infra (VPS, AWS, GCP) or on ours as a managed service. Zero dependency on Intercom, Drift, Zendesk or other opaque chatbot SaaS.
Three reasons. (1) Cost: standard chatbot SaaS start at 200-500€/month on a professional plan and scale with conversations; a custom bot has Claude token cost (cents per conversation) + infra (€10-30/month VPS). (2) Customization: a custom bot can connect to ANY endpoint of yours, not just the ones the SaaS provides as an integration. (3) Data: the conversations stay on your infra, not on a third party's servers — useful for regulated cases (clinics, accounting firms, legal).
It depends on complexity. A simple bot (FAQ with RAG + lead delivery to CRM): low range, weeks of implementation. A bot connected to 4-5 internal systems with complex tool use and guardrails: mid-to-high range, 4-8 weeks. Recurring operating cost: Claude tokens (cents per conversation with aggressive prompt caching) + infra (€10-30/month). Fixed price after a 30-min assessment, no surprises.
The base model can hallucinate. We mitigate it with: (1) strict RAG — the bot only answers with info from your indexed docs and cites the source; (2) tool use for real-time data (Holded, CRM) instead of making things up; (3) explicit guardrails — the bot knows which topics escalate to a human (prices, timelines, special cases) and does so; (4) testing with real edge cases before production. Even so it isn't 100%, which is why there's always 'escalate to a human' as a visible button.
Yes. Claude works natively in Spanish, English, French, Italian, German, Portuguese and others. The bot detects the user's language and replies in the same one. For your Spanish knowledge base, we also index translations when it's relevant (typically only if you have international clients).
Yes, with considerations. Conversations containing personal data are logged in YOUR database (not in third-party SaaS). The LLM processing goes through Anthropic (AWS, EU if you configure Bedrock). We recommend: an explicit privacy notice at the bot's opening, an option to delete history, anonymization of old logs. For heavily regulated sectors (clinics with health data), we evaluate case by case whether Claude on Bedrock EU or Vertex AI Europe fits your DPA.
Yes, using the official WhatsApp Business API (Meta Cloud API or 360dialog as a BSP). It requires a dedicated number and verification. We set it up as part of the package if you include it. For small volumes (<1000 conversations/month) the WhatsApp Business API has a free tier; for high volume there's a per-conversation cost from Meta + our Claude token cost.
The bot lives in your Git repo with your team or ours as maintainers. Changes to the knowledge base (prices, new FAQs) are made by editing the indexed docs — the bot reindexes on its own. Structural changes (new tool, new integration) are dev work. We offer monthly or on-demand maintenance depending on your model.
Related: Custom MCP servers · Holded Claude integration
A 30-minute call. We tell you whether your case fits one of the 5 patterns or needs custom work, how much it costs and how long it takes.
