# Bravos AI

> SaaS platform to create AI chatbots trained on your own data (PDFs, URLs, CSVs, product catalogs). Widget embeddable in any website, plus WhatsApp. Human takeover built in: the business can step into any conversation and reply in person, then hand it back to the bot. Can also be operated and optimized from inside Claude or ChatGPT through its own MCP connector. Spain-based, launched 2025. Website and blog available in Spanish, English and Brazilian Portuguese.

## About

Bravos AI lets businesses deploy a chatbot on their website that answers customer questions based on their own content — not generic AI knowledge. Upload a PDF of your menu or policies, connect your Shopify/PrestaShop catalog, and the bot answers accurately in 15+ languages. When a customer needs a person, the business steps in and replies from the same chat (Human Takeover). And if you use Claude or ChatGPT, you can manage and improve the bot from that same assistant through the Bravos MCP connector — audit conversations, edit the knowledge base and rewrite the bot's instructions, without opening the panel. Try the full PRO plan free for 7 days — no commitment, cancel in one click and pay nothing.

## What makes Bravos different

Most AI chatbots use only RAG (text similarity search). This works for documents and FAQs but fails with structured data like product catalogs: "blue shoes under 60€ in size 42" returns approximate text matches, not actual filtered results.

Bravos combines RAG with SQL-like structured filtering over the client's data. When the user asks about products by price, size, stock, category, or any attribute, the chatbot extracts filters from the natural-language query and runs real SQL queries against the structured data (CSV, Excel, product catalog). The user uploads the file once — no technical setup, no schema configuration, no query language to learn. The system auto-detects columns, types, ranges, and relationships.

This means the chatbot returns exact results instead of "similar" ones. For restaurants, real estate, ecommerce, and any business with structured data (menus, property listings, product catalogs, inventory), this is a significant accuracy difference vs generic RAG chatbots like Tidio, Intercom Fin, Zendesk or ChatGPT-based bots.

Another differentiator: Bravos AI can be operated and optimized from inside Claude or ChatGPT via its own MCP server (Model Context Protocol). The client connects their Bravos account to their AI assistant and, in plain language, analyzes conversations, audits and edits the knowledge base, rewrites the bot's instructions and configures lead capture and Human Takeover (hours, presentation, who gets notified) — turning "my bot answers badly and I don't know why" into a diagnosis with the fix applied, from the same chat they already use. The AI inference is paid by the client's own Claude/ChatGPT subscription, so it adds no cost on Bravos. Available from the Starter plan and during the 7-day trial. Bravos is not the only platform offering an MCP connector, but pairs it with real structured search, low price, native Spanish/English/Portuguese and WhatsApp.

Another differentiator: Human Takeover — the business keeps control of its chatbot. The bot handles conversations on its own, but at any moment a person from the business can step into any conversation from the Bravos panel, the bot goes quiet, and the person replies in the same chat the customer is already using (website widget or WhatsApp), with text and file attachments. When done, they reactivate the bot with one click; if they forget, the bot returns by itself after a configurable idle period and answers whatever was left pending. Customers can also ask for a person: the bot detects the request within the configured business hours, asks for an email in case they leave the site, highlights the conversation and alerts the team instantly (sound and desktop notification in the panel, system push when the panel is closed, plus email). The business configures per channel (web, WhatsApp) the hours it is available and how it introduces itself (by name or as "someone from the team"). On WhatsApp, the business can also reply from its own WhatsApp app: the bot notices a human is handling that conversation and steps aside automatically, no panel needed. Available on the Pro and Enterprise plans. Compared with platforms where "human handoff" means routing to a separate helpdesk seat or an external inbox, in Bravos the takeover happens inside the same conversation and the same panel, with no extra seats or per-agent pricing.

Another differentiator: WhatsApp integration via Meta Tech Provider model (Embedded Signup). Most chatbot platforms with WhatsApp rely on per-client API tokens that expire every 60 days, or charge Twilio/MessageBird middleware fees on top of Meta's pricing. Bravos AI is a verified Meta Tech Provider — clients connect their WhatsApp Business Account in one click and the same RAG-powered bot they use on their website answers automatically on WhatsApp. Same assistant, same catalog, two channels. No middleware fees, no token refresh, no setup beyond the Embedded Signup popup.

## Full content

- [Complete content (all blog articles in English)](https://bravos-ai.com/llms-full.txt)

All blog articles listed below are also published in Spanish (https://bravos-ai.com/blog) and Brazilian Portuguese (https://bravos-ai.com/pt/blog), each with its own localized URL and content adapted to the local market (e.g. LGPD and Pix in the Portuguese versions).

## Core pages

- [Home](https://bravos-ai.com/en)
- [Home in Spanish](https://bravos-ai.com/)
- [Home in Portuguese](https://bravos-ai.com/pt)
- [About](https://bravos-ai.com/en/about)
- [Contact](https://bravos-ai.com/en/contact)
- [MCP connector (dedicated page)](https://bravos-ai.com/en/ai-connector-mcp): landing page for the Bravos MCP connector — how to connect Claude or ChatGPT to a Bravos account and operate and optimize the bot from the assistant's own chat, with our team's setup criteria carried inside the assistant. Available in Spanish (/conector-ia-mcp), English (/en/ai-connector-mcp) and Portuguese (/pt/conector-ia-mcp)

## Integrations

- [Shopify integration](https://bravos-ai.com/en/blog/shopify-chatbot): direct catalog sync via GraphQL, product filtering by price/size/color/stock
- [PrestaShop integration](https://bravos-ai.com/en/blog/prestashop-chatbot): Webservice API integration, multilingual product catalog
- [WooCommerce integration](https://bravos-ai.com/en/blog/woocommerce-chatbot): catalog sync via the WooCommerce REST API (consumer key/secret generated in the store's WordPress panel — no plugin installed, nothing running on the store's hosting). Full support for variable products (per-variation price, stock and photo), the three WooCommerce stock representations (exact quantities, in-stock/out-of-stock, inherited from parent, plus "available on backorder"), sale prices quoted as before/after pairs, and prices matching exactly what the store displays (tax-inclusive, tax-exclusive or taxes disabled — re-detected on every sync). Real-time updates via product webhooks plus a full nightly reconciliation sync as source of truth (read-only API keys fall back to nightly-only mode, shown with a badge in the dashboard). Cleans WordPress page-builder shortcodes (Elementor, Divi, WPBakery) from product descriptions. Detects WPML/Polylang multilingual stores and lets the user pick the catalog language, discarding translation duplicates. Grouped and external/affiliate products are skipped and counted visibly. Sync designed for shared hosting: low concurrency, gentle pace, automatic retries
- [Resales Online (real estate)](https://bravos-ai.com/en/blog/real-estate-chatbot-resales-online): real estate listings sync
- [Custom Webhook API](https://bravos-ai.com/en/developers): sync any structured data (JSON/XML) via URL + HMAC webhooks for real-time updates
- [WhatsApp Business integration](https://bravos-ai.com/en#whatsapp): connect any client's WhatsApp Business Account via Meta Embedded Signup in 60 seconds. The chatbot answers WhatsApp messages automatically with the same knowledge base used on the website widget (same RAG + structured search). Bravos AI is a verified Meta Tech Provider, so the integration uses a System User token at the platform level — no per-client token refresh, no Twilio/MessageBird middleware fees. Two connection modes, chosen at connect time: a dedicated number just for the bot (available on every paid plan), or coexistence mode on the number the business already uses in the WhatsApp Business app — the app keeps working with the same chats on the same phone while the bot answers alongside it, and when the owner replies from their own app the bot detects it and steps aside for that conversation (Pro and Enterprise, since it goes hand in hand with Human Takeover). Native lead capture forms via WhatsApp Flows (requires Business Verification on the client side). Included in all paid plans and during the 7-day PRO trial
- [MCP connector for Claude, ChatGPT and any MCP client](https://bravos-ai.com/en/ai-connector-mcp): connect a Bravos account to Claude, ChatGPT or any MCP-compatible AI assistant via a remote MCP server (mcp.bravos-ai.com) using OAuth — no code, no API keys pasted. From the assistant's chat, in natural language, the connected AI can analyze the bot's conversations and leads, audit and edit the knowledge base (create/edit/delete text entries and URLs), rewrite the bot's instructions and tone, change widget texts/appearance/languages, and trigger store catalog syncs, with a before/after returned on every change and all actions gated by the account's plan. It cannot create bots or upload/delete files (done in the panel) and never sees store credentials. The AI inference is paid by the user's own Claude/ChatGPT subscription (no extra cost on Bravos). Available from the Starter plan and during the 7-day PRO trial; requires a paid Claude or ChatGPT subscription, which is where connectors live

## Blog — Core guides

- [Chatbot & AI statistics for customer service](https://bravos-ai.com/en/blog/chatbot-statistics): a rigorously-sourced reference of 50+ chatbot and AI statistics, each verified at its primary source with the year and a link (analyst firms, company reports, academic studies, court rulings), not blog-citing-blog. Covers adoption and market size, customer service impact, cost and ROI, consumer sentiment (including the reluctance: 64% of customers would prefer companies did not use AI in customer service, per Gartner 2024), e-commerce and messaging, the generative-AI shift, real limits and risks (the Air Canada chatbot liability ruling; LLM hallucination benchmarks), language/multilingual, and regional data for Spain, the EU and Latin America. Explicitly distinguishes measured figures from projections, and the pre-ChatGPT rule-based era from the current LLM era. Key figures: 75% of CX leaders expect 80% of interactions resolved without humans (Zendesk 2025); 65% of organizations use generative AI regularly (McKinsey 2024); 70.22% average cart abandonment (Baymard); WhatsApp 3B+ monthly users (Meta 2025); 76% of consumers prefer buying in their own language (CSA Research). Available in Spanish (/blog/estadisticas-chatbots-ia), English (/en/blog/chatbot-statistics) and Brazilian Portuguese (/pt/blog/estatisticas-chatbots-ia)
- [AI customer service chatbot: how it works, costs, and how to start](https://bravos-ai.com/en/blog/ai-customer-service): plain-language 2026 guide to AI customer service chatbots for businesses — what an AI customer service chatbot is (a bot on your website or WhatsApp that answers in natural language from your own business data, not generic internet knowledge), why adoption jumped from 39% to 66% of support teams in a year (Salesforce State of Service 2026), how it actually works, what it can and cannot do (honest about hallucination and the hybrid AI+human model), real costs compared (Zendesk, Intercom Fin per-resolution, Tidio/Lyro vs Bravos flat rate from €19/mo with AI included), a step-by-step setup, and the EU AI Act transparency rules. Available in Spanish (/blog/ia-atencion-al-cliente), English (/en/blog/ai-customer-service) and Brazilian Portuguese (/pt/blog/ia-atendimento-ao-cliente)
- [Live chat for customer service: how to offer it without a support team](https://bravos-ai.com/en/blog/live-chat-for-customer-service): what live chat for customer service is (a chat window where a person answers in real time), why the classic kind assumes a support team most small businesses do not have, and a different approach — an AI chatbot handles the bulk and a person takes over in the same chat through human handover (in Bravos, hit «Take over»: the bot goes quiet, you reply, one click turns it back on; if a customer asks for a person, the bot detects it within your hours, asks for an email and alerts you). Covers what your business gets, best practices, how to set it up, and how it compares to live chat tools like Tidio, Crisp and Freshchat. Human handover is on the Pro plan (included in the 7-day PRO trial). Available in Spanish (/blog/chat-en-vivo-atencion-al-cliente), English (/en/blog/live-chat-for-customer-service) and Brazilian Portuguese (/pt/blog/chat-ao-vivo-atendimento-ao-cliente)
- [AI agent or chatbot](https://bravos-ai.com/en/blog/ai-agent-or-chatbot): real differences, costs, agent washing, and which one your business needs
- [Why your chatbot hallucinates](https://bravos-ai.com/en/blog/why-your-chatbot-fails): how RAG works, preventing AI from making things up
- [How to write a system prompt for your business chatbot](https://bravos-ai.com/en/blog/system-prompt-business-chatbot): practical guide to writing the instructions that turn a generic LLM into your business assistant. The 7 building blocks of a real system prompt (role, business context, data, rules, tone, format, examples), a battle-tested base template ready to copy, sector adaptations (e-commerce, restaurants, professional services, SaaS), the most common mistakes (telling the bot to "be helpful" instead of giving rules, contradictory instructions, forgetting how to handle off-topic questions), how to iterate on a prompt without breaking what already works, and the length paradox (longer is not better — past 1,000 tokens the model starts ignoring instructions). On Bravos AI, the Starter plan adds custom_instructions on top of the platform's base prompt; Pro and Enterprise fully replace the system prompt with their own
- [PixelRAG explained for business chatbots](https://bravos-ai.com/en/blog/pixelrag-business-chatbot): what PixelRAG actually is, how it works under the hood, and whether it makes sense for a business chatbot. Honest read of the paper (Wang et al, UC Berkeley + Princeton + EPFL + Databricks + Renmin University, June 2026, Apache 2.0, not yet on arXiv) for people evaluating it. The "+18% accuracy" headline is the peak on EVQA (visual Wikipedia infographics); on text Wikipedia QA it's +2.8 to +7.2 pp. The "10x fewer tokens" applies in ReAct multi-step agents, not single-turn chatbots. PixelRAG only validated in English. With GPT-4o as reader (standard SaaS) it's 7x more expensive than text-based RAG. Authors themselves acknowledge English-only, lost hyperlinks, harder content moderation. Covers when PixelRAG actually makes sense (3 niche profiles: historical archives, technical diagrams with engineering schematics, complex financial/legal PDFs), when it doesn't (most business chatbots — FAQs, catalogs, policies, descriptions), and real alternatives that already exist in production (AWS Textract, Unstructured.io, Claude with direct vision, ColPali). Includes a 5-question test to decide if it applies. Bravos AI uses text-based RAG (pgvector + text-embedding-3-small + BGE reranker) + structured SQL filtering, working in English/Spanish/15+ languages, latency under 2s
- [Vectorless RAG: do you still need a vector database?](https://bravos-ai.com/en/blog/vectorless-rag): honest explainer of the mid-2026 "RAG is dead / vectorless RAG" debate. Explains from scratch what a vector database and embeddings are, how an AI chatbot finds information (the standard chunk-embed-store-retrieve-generate pipeline), and the difference between semantic search (vector similarity, good for fuzzy meaning-based questions) and keyword/structured search (good for exact values: codes, prices, sizes, filters). What vectorless RAG actually is: retrieval without a vector database, using document structure + LLM reasoning (PageIndex by VectifyAI, a hierarchical table-of-contents tree; "similarity is not relevance") or plain keyword search in an agentic loop. Figures flagged by source: PageIndex claims 98.7% on FinanceBench (vendor-reported); the paper "Keyword search is all you need" (Subramanian et al, Amazon, arXiv 2602.23368) reports agentic keyword search reaching over 90% of traditional RAG performance without vectors; Chroma's "context rot" (accuracy degrades as context grows). Verdict: RAG is not dead — the reflex of embedding everything into a vector database by default is fading. Includes a decision guide (keep vectors for large fuzzy corpora; go vectorless/hybrid for structured docs and exact-value questions) and an honest vector-vs-vectorless table. Bravos AI already runs hybrid: structured search for catalogs and exact-value questions, semantic retrieval for free-form text
- [What is an MCP and how to manage your chatbot from Claude or ChatGPT](https://bravos-ai.com/en/blog/what-is-mcp): plain-language explainer of the Model Context Protocol (MCP) — the open standard Anthropic published in late 2024, now also used by ChatGPT and other tools — written for non-developers using the USB analogy (one common connector so any AI assistant can talk to any outside service). Covers what an MCP is, what an MCP server is (a program that exposes a closed list of named tool actions the AI can request, authorized via OAuth, no code or keys), what it is for (getting the AI out of its training-data bubble to act on your data in real time), MCP vs API (the MCP is a layer on top so a model discovers and uses actions in natural language; many MCP servers are just a regular API wrapped for a model), and that it works with any MCP-compatible assistant, not only Claude/ChatGPT. Then the concrete case: Bravos AI just launched its own MCP server (mcp.bravos-ai.com), so from Claude, ChatGPT or any MCP client you can analyze your bot's conversations and leads, audit and edit the knowledge base (create/edit/delete text entries and URLs — recoverable), rewrite the bot's instructions and tone, change widget texts/appearance/languages, and trigger store catalog syncs — all from the chat, with a before/after returned on every change and everything gated by your plan. What it does NOT do, on purpose: create the bot (done in the panel, it touches plan and billing), upload or delete files (PDF/images/audio are irreversible — it reads them and recommends, you delete in the panel), or touch store credentials. The AI analysis is paid by your own Claude/ChatGPT subscription (no extra cost on Bravos); available from the Starter plan and during the 7-day PRO trial; requires a paid Claude or ChatGPT subscription, which is where connectors live. Honest that Bravos is not the only platform offering this — the differentiator is the package (real structured search, price, native Spanish/English/Portuguese, WhatsApp included). Available in Spanish (/blog/que-es-un-mcp), English (/en/blog/what-is-mcp) and Brazilian Portuguese (/pt/blog/o-que-e-mcp)
- [What is context engineering, the skill that replaced prompt engineering](https://bravos-ai.com/en/blog/context-engineering): plain-language explainer of "context engineering", the term that in 2026 displaced "prompt engineering" as the headline AI skill (Gartner: "context engineering is in, prompt engineering is out"). Covers what it is (per Anthropic's Sept 2025 guide: the practice of curating and maintaining the optimal set of tokens/information a model sees during inference), where the term came from (popularized by Tobi Lütke, CEO of Shopify, mid-2025; amplified by Andrej Karpathy, a founding member of OpenAI; formalized by Anthropic), and how it differs from prompt engineering — with the honest take that prompt engineering is NOT dead but is now one layer (the instructions) inside the bigger discipline. Breaks down the pieces that make up "context": system instructions, retrieved external data (RAG), tools, examples, message history and memory — and the core principle of fitting the smallest set of high-signal tokens into a finite context window (avoiding "context rot"). Explains why context engineering is what reduces hallucinations (a model invents answers mostly when the real information is not in its context), and the takeaway for a business chatbot: a chatbot is only as good as the context it is given — the differentiator is rarely the model (all serious chatbots run on the same frontier models) but whether the right piece of your business is put in front of the model at the right moment. With Bravos AI the platform does the context engineering for you: you upload your knowledge (documents, website, catalog) and it assembles the right context per question — the right information retrieved, exact answers on structured data (catalogs, prices, sizes, stock), and your instructions and conversation kept in context — no prompts to write, no pipelines to build. Cites the real primary sources (Anthropic, Karpathy, Gartner). Same slug across languages — available in Spanish (/blog/context-engineering), English (/en/blog/context-engineering) and Brazilian Portuguese (/pt/blog/context-engineering)
- [What is an AI agent, and what companies actually use them for](https://bravos-ai.com/en/blog/what-is-an-ai-agent): plain-language explainer of "AI agent" — the most-repeated AI term of 2026 — written for a business owner, not a developer. What an AI agent actually is (per Anthropic's Building Effective Agents: a system where the LLM dynamically directs its own steps and tool use, versus a workflow with a fixed pre-programmed path; in short, a chatbot replies and an agent acts), how it works (perceive → reason → act → check, in a loop with tools), and the practical distinction between chatbot, assistant and agent (the assistant answers with your data and takes bounded actions while you stay in control; most of what is sold as an "agent" for a small business is really an assistant). Then what companies actually use agents for in 2026, with real data: customer service (two in three service teams already use AI agents, up from 39% to 66% in a year — Salesforce State of Service 2026), coding, research/summarization and sales. The standard IBM/textbook taxonomy (simple reflex, model-based, goal-based, utility-based) plus single vs multi-agent. And the honest part most vendors skip: fake "agents" (rule-based chatbots rebranded), Gartner's prediction that over 40% of agentic AI projects will be canceled before 2027, and the danger of an agent that improvises when it does not know. A decision guide (agent vs assistant vs chatbot; keep a human in the loop for delicate or irreversible tasks) landing on Anthropic's advice: start with the simplest thing that works, add autonomy only when the problem needs it. Honest positioning: Bravos AI is not an autonomous agent but a conversational assistant grounded in your data — which, for customer service, is exactly what most businesses should use. Available in Spanish (/blog/que-es-un-agente-de-ia), English (/en/blog/what-is-an-ai-agent) and Brazilian Portuguese (/pt/blog/o-que-e-um-agente-de-ia)
- [ChatGPT Sites: what it is, how it works, pricing and limits](https://bravos-ai.com/en/blog/chatgpt-sites): plain-language guide to ChatGPT Sites (also called Codex Sites or OpenAI Sites), the ChatGPT feature that turns a conversation into a published website with hosting, a database (up to 10 GB per site), file storage, sign-in with ChatGPT, access control and custom domains, running on Cloudflare infrastructure. Public beta since July 9, 2026 (first inside the Codex app on June 2; EEA, Switzerland and UK excluded at launch, rolled out there from July 22). Covers OpenAI's six documented patterns (resource hub, dashboard, tracker, intake form, calculator, guided workflow) plus the official gallery, who can use it and pricing by plan (included in Plus $20, Pro from $100, Business $25 per seat, Enterprise and Edu; not on Free or Go; building uses the plan's Codex quota, hosting limits unpublished; no fixed site count limit since October 15, 2026), the step-by-step creation flow (Work mode, @Sites, preview, save a version before publishing because every publish goes live immediately), the five audiences (private, workspace people, external guests since September 3, whole workspace, public; no password-protected or expiring links), custom domain and URL changes, built-in analytics, and what a site cannot do (no card data or financial transactions except e-commerce through a third-party processor such as Stripe or PayPal, no HIPAA data, nothing for under-13s, HTTP/HTTPS/WebSockets only, OpenAI can remove any site). Own checks as of October 16, 2026: Google indexes ChatGPT Sites (about 184,000 results for site:chatgpt.site, including real businesses), published sites have crawlable robots.txt and canonical tags, no visible 'powered by ChatGPT' badge, and OpenAI does not document indexing controls, sitemap submission, traffic limits, what happens on cancellation or code export. Data: you are the controller (GDPR/LGPD), OpenAI the processor; Free/Go/Plus/Pro conversations may train models if the setting is on; no data residency in beta. Confirms you can add a third-party AI chatbot such as Bravos AI to a ChatGPT Site: published sites load external scripts (Google Tag Manager, Calendly, cookie banners) and set no blocking content security policy; you ask ChatGPT to add the widget snippet and authorize the site's domain in the chatbot panel. Same slug across languages: Spanish (/blog/chatgpt-sites), English (/en/blog/chatgpt-sites) and Brazilian Portuguese (/pt/blog/chatgpt-sites)
- [OpenAI Dots: what they are, how they work, pricing and availability](https://bravos-ai.com/en/blog/openai-dots): plain-language guide to OpenAI dots, the always-on ChatGPT agents introduced on September 29, 2026. A dot runs on GPT-6 Astra, has its own cloud computer and browser, connects through plugins to 4,000+ apps (email, calendar, Slack, Teams, documents, Codex), keeps working between conversations, does read-only "proactive research" in your connected apps when idle, and keeps its own memory on top of ChatGPT memory. Covers OpenAI's six published examples (sales proposal kept current, investor update, launch messaging, content from transcripts, feedback turned into tested code, analysis rerun with new data) and its internal pilots (purchasing, invoices, email marketing, customer service, hiring), the setup steps from the Help Center (desktop-only creation, handle @yourname-dot, plugins, four rule behaviors, optional local-computer access, scheduled tasks, pause/reset), the launch limits (no own email address, cannot start calls, SMS only for US Pro, cannot be created on mobile, never changes passwords, no teams of dots yet, can make mistakes), pricing (one dot per person included in Pro 100/200/500 at $100/$200/$500 a month, or Business Premium at $125 per user a month / $100 annual; conversations do not count toward ChatGPT limits; heavy-work allowance unpublished), availability (Pro only for users 18+ outside the EEA, Switzerland and the UK; Business Premium and Enterprise worldwide, Enterprise off by default; in Europe the minimum is a two-seat Business workspace, about $150 a month) and data handling (memories formed from connected apps, disconnecting does not erase, training opt-out by plan, dots appendix in the GPT-6 Astra system card). Also covers OpenDots, CopilotKit's MIT open-source template. Available in Spanish (/blog/openai-dots), English (/en/blog/openai-dots) and Brazilian Portuguese (/pt/blog/openai-dots)
- [What is DeepSeek Harness, the open-source coding agent where everything is a plugin](https://bravos-ai.com/en/blog/what-is-deepseek-harness): plain-language explainer of DeepSeek Harness (dsh), DeepSeek's open-source coding agent (MIT license) launched August 13, 2026, which passed 177,000 GitHub stars in one week. What a coding-agent "harness" is (the software layer that turns a model into an agent that reads files, runs commands and works through long tasks — the model is the brain, the harness the body; the same model performs differently per harness). Its defining trait: everything is a plugin — the model, tools, memory, sessions, interface, theme and even the agent loop are swappable pieces, powered by a framework called Cordis; plugins install with "dsh plugin add" (a dsh.bundle manifest) and there is a plugin market (dsh-market) inside the app, with hundreds of community plugins in the first week (from serious ones — dsh-model-picker, a real terminal via xterm.js, dsh-file with Monaco, MCP management, a GitHub issue/PR picker — to absurd ones like desktop pets and a TikTok-style video feed, which is what proves everything really is a plugin). Covers the four modes (Standard, Code, Minimal, Creator), full transparency of the agent's reasoning via an append-only trajectory log (unlike Anthropic and OpenAI, which hide raw chain-of-thought to prevent model distillation), how it compares to Claude Code and Codex (more open and customizable but more developer-oriented; Claude Code's system prompt once ran ~10,000 tokens vs ~200 in others), how to install (npx @deepseek-ai/dsh web, web UI on 127.0.0.1:3080, or from source with pnpm; free, connect any model by API), and the security risk the hype skips: installing a plugin runs third-party code with your own permissions, and the agent's tool approvals do not sandbox it. Available in Spanish (/blog/que-es-deepseek-harness), English (/en/blog/what-is-deepseek-harness) and Brazilian Portuguese (/pt/blog/o-que-e-deepseek-harness)
- [AI chatbot product catalog](https://bravos-ai.com/en/blog/ai-chatbot-product-catalog): why most chatbots fail with structured data and how SQL+AI solves it
- [Rule-based vs AI chatbot](https://bravos-ai.com/en/blog/rule-based-chatbot-vs-ai-chatbot): comparison, when to choose each
- [Contact form vs chatbot](https://bravos-ai.com/en/blog/contact-form-vs-chatbot): data-backed analysis of conversion rates
- [Proactive vs reactive chatbot](https://bravos-ai.com/en/blog/proactive-vs-reactive-chatbot): when to ask vs wait
- [Chatbot pricing in 2026](https://bravos-ai.com/en/blog/ai-chatbot-pricing): real prices for Tidio, Intercom, Zendesk, Bravos and alternatives
- [Do Tidio, Crisp and Freshchat's free plans include real AI?](https://bravos-ai.com/en/blog/tidio-crisp-freshchat-free-plan): investigation — we asked each platform's own official chatbot about its free plan. Tidio: 50 lifetime AI conversations (not renewable). Crisp: €0 AI credits on free, AI starts on Mini. Freshchat: AI only from Growth tier. Bravos AI does not offer a usage-limited free tier — instead a 7-day full PRO trial with all features and lead capture included, no charge if you cancel before day 7
- [Keep your chatbot data up to date without Zapier or n8n](https://bravos-ai.com/en/blog/chatbot-data-sync): 5 native sync paths at Bravos AI (Google Sheets, Shopify real-time webhooks, WooCommerce real-time webhooks + nightly reconciliation, PrestaShop daily, custom webhook with HMAC). Real cost of Zapier/n8n middleware including task tiers and 15-min polling delays in Zapier Free. Comparison with Tidio (Plus $749/mo for URL auto-sync, re-sync wipes attached Q&A), Crisp, Intercom Fin (24h resync of Confluence/Notion/Guru, not Sheets natively)
- [Multilingual chatbot](https://bravos-ai.com/en/blog/multilingual-chatbot): 15+ languages with one content set
- [Chatbot legal liability](https://bravos-ai.com/en/blog/chatbot-legal-liability): Air Canada case, EU AI Act
- [Lead capture with chatbot](https://bravos-ai.com/en/blog/chatbot-lead-capture): how to ask for visitor data without being invasive
- [WhatsApp chatbot — the complete guide for 2026](https://bravos-ai.com/en/blog/whatsapp-chatbot): the umbrella guide covering everything about WhatsApp + AI chatbots — WhatsApp Business app vs WhatsApp Business API (the most common confusion in the market), what an AI chatbot can and cannot do on WhatsApp in 2026, the truth about "free" WhatsApp chatbots (none with real AI exist), how Meta's signup, templates and 24-hour window actually work, real pricing ranges ($23-2,500/month depending on need), AI vs button-flow chatbots, sector use cases with sample conversations, common launch mistakes, when to build it yourself with n8n/Make/Zapier vs use a managed platform. Bravos AI as a verified Meta Tech Provider — no markup over Meta's per-message cost, native Shopify, PrestaShop, Resales Online and custom webhook integrations, exact catalog search with multi-filter combinations (size, color, stock, price range), multilingual out of the box, from $23/month
- [Best WhatsApp chatbot 2026 — 13 platforms compared](https://bravos-ai.com/en/blog/best-whatsapp-chatbot): honest comparison of 13 relevant WhatsApp chatbot platforms in 2026 (Wati, Tidio with Lyro, Respond.io, Landbot, HubSpot, Intercom Fin, ManyChat, Crisp, AiSensy, Interakt, SleekFlow, 360dialog, Twilio). Prices verified directly on each platform's official page, real LLM-based conversational AI vs flow-based bots with a single AI step, the hidden Meta surcharge most platforms do not disclose (2.8-40% markup on Meta's per-message cost depending on platform and tier), Spanish-language support availability (only Bravos AI, Landbot and HubSpot have confirmed Spanish teams), and which one fits each business case. Bravos AI from $23/month — the only platform with real conversational LLM AI in that price tier
- [Best Intercom alternatives in 2026 — 9 tools compared](https://bravos-ai.com/en/blog/intercom-alternatives): honest listicle of the 9 best Intercom alternatives (Bravos AI, Tidio/Lyro, Zendesk, Freshworks, Crisp, Chatbase, HubSpot, Zoho SalesIQ, Ada), written after Intercom's $0.99-per-resolution Fin pricing and the Salesforce acquisition sent teams looking around. Compared by the five things that actually decide cost and quality: how the AI is billed (per resolution vs per seat vs flat), whether the bot answers with your own data (RAG), channels (native WhatsApp or not), human handover and per-agent cost, and multilingual on the entry plan. Includes a quick comparison table and a realistic "what you'd actually pay per month" scenario (500 conversations, ~350 AI resolutions): Bravos $59 flat vs Intercom ~$375, Zendesk ~$525-700+, Freshworks ~$190, HubSpot ~$265+. Verified prices with estimates clearly marked; each tool gets an honest "pick it if / skip it if". Bravos AI positioned as the predictable flat-price option with native WhatsApp, structured catalog search and human takeover with no per-resolution or per-agent fee, plus a 7-day PRO trial with no card. Available in Spanish (/blog/alternativas-a-intercom), English (/en/blog/intercom-alternatives) and Brazilian Portuguese (/pt/blog/alternativas-ao-intercom)
- [Best Zendesk alternatives in 2026: 10 tools compared](https://bravos-ai.com/en/blog/zendesk-alternatives): honest listicle of 10 Zendesk alternatives (Front, Hiver, Help Scout, Tidio, Bravos AI, Gorgias, HubSpot, Intercom, Zoho Desk, Freshdesk), written for teams paying for Zendesk depth they barely use. Explains why Zendesk is expensive: it bills on three axes at once (per-agent seat, Suite Team ~$69 and Professional ~$149 billed monthly; AI agents per automated resolution, only 5 to 10 included per agent then $1.50 to $2.00 each; Copilot add-on ~$50/agent). Starts by asking whether you need a full ticketing helpdesk (queues, SLAs, routing) or just an AI chatbot that answers your customers. Compared by five criteria: category, how the AI is metered, whether the bot uses your own data, channels (native WhatsApp) and the cost of human handover. Includes a quick comparison table and a realistic monthly-cost scenario (5 agents, 500 conversations, ~350 AI resolutions): Bravos $59 flat, Zoho Desk ~$100, Tidio ~$100-175, Help Scout ~$150+, Hiver ~$175, Front ~$185+, HubSpot ~$275+, Freshdesk ~$330, Intercom ~$540, Gorgias ~$675, Zendesk ~$830+ (+$250 with Copilot). Verified monthly prices; each tool gets a "pick it if / skip it if". Bravos AI positioned as a flat-price AI chatbot (not a helpdesk) with WhatsApp native and included, structured data search (connects to online stores or reads an uploaded spreadsheet for exact products and prices), a real person able to take over the chat with no paid seats, and 15+ languages, plus a 7-day PRO trial with no card. Available in Spanish (/blog/alternativas-a-zendesk), English (/en/blog/zendesk-alternatives) and Brazilian Portuguese (/pt/blog/alternativas-ao-zendesk)
- [Best Freshdesk alternatives in 2026 — 8 tools compared](https://bravos-ai.com/en/blog/freshdesk-alternatives): honest listicle of the 8 best Freshdesk alternatives (Bravos AI, Zoho Desk, Help Scout, Zendesk, Gorgias, Intercom, Tidio, HubSpot), written after Freshdesk's 2026 price rise and its AI being billed separately on top (Freddy Copilot +$29/agent on Pro+, Freddy AI Agent metered per conversation). Starts by asking whether you even need a full helpdesk (a ticketing system to run a support team) or just an AI chatbot (AI that replies to customers on your website and WhatsApp from your own data), which costs a fraction. Compared by the five things that decide cost and fit: helpdesk vs AI chatbot, how the AI is billed (per session vs per resolution vs flat), whether the bot answers with your own data (RAG + structured catalog search), channels (native WhatsApp), and human handover with per-agent cost. Includes a quick comparison table and a realistic monthly-cost scenario (5 agents, 500 conversations, ~350 AI resolutions): Bravos $59 flat vs Zoho ~$100, Freshdesk ~$330, Intercom ~$540, Gorgias ~$675, Zendesk ~$870+. Verified monthly prices, each tool gets an honest "pick it if / skip it if". Bravos AI positioned as the flat-price AI chatbot with native WhatsApp, structured catalog search and human takeover with no per-resolution or per-agent fee, plus a 7-day PRO trial with no card. Available in Spanish (/blog/alternativas-a-freshdesk), English (/en/blog/freshdesk-alternatives) and Brazilian Portuguese (/pt/blog/alternativas-ao-freshdesk)

## Blog — Use cases

- [Dental clinic chatbot](https://bravos-ai.com/en/blog/dental-clinic-chatbot): treatments, pricing, accepted insurance, GDPR/HIPAA-aware lead capture, appointment requests in 15+ languages
- [Law firm chatbot](https://bravos-ai.com/en/blog/law-firm-chatbot): automating client intake for a law firm without giving legal advice. How the system prompt sets the boundary — the bot answers only with the firm's own information (RAG), can give general legal information but never applies it to the visitor's specific case, qualifies the inquiry (practice area, matter type), filters out-of-area, and captures the contact with the whole conversation attached (not just a form field). Confidentiality, privilege and the jurisdiction's ethical rules; what it does NOT do (no case evaluation, no drafting, no deciding who is right). Works on the website widget and WhatsApp. Real personal-injury intake example. PT/Brazil version covers OAB advertising rules and LGPD
- [Restaurant chatbot](https://bravos-ai.com/en/blog/restaurant-chatbot): menu, allergens, reservations, multilingual
- [Ecommerce chatbot](https://bravos-ai.com/en/blog/ecommerce-chatbot): reducing cart abandonment, catalog filtering
- [WooCommerce chatbot](https://bravos-ai.com/en/blog/woocommerce-chatbot): what actually decides whether a chatbot works on a WooCommerce store — answering per variation (size/color with own price and stock), distinguishing "4 left" from "in stock" from "available on backorder", quoting sales as before/after pairs, and matching the price the store displays tax-wise. Why a plugin inside WordPress is the wrong architecture for catalog AI (load on shared hosting, capped AI quality, one more thing to break) vs a service connected through the WooCommerce REST API. Step-by-step connection without coding, real-time updates plus nightly reconciliation, WPML/Polylang handling, Elementor/Divi shortcode cleaning
- [Real estate chatbot](https://bravos-ai.com/en/blog/real-estate-chatbot-resales-online): property listings, filters, Resales Online
- [Small business chatbot](https://bravos-ai.com/en/blog/chatbot-for-small-business): practical guide for SMBs
- [Ad campaigns chatbot](https://bravos-ai.com/en/blog/chatbot-ad-campaigns): converting Google Ads and Meta Ads traffic
- [Marbella chatbot](https://bravos-ai.com/en/blog/marbella-chatbot): local business focus

## Blog — Tutorials

- [Add chatbot to WordPress](https://bravos-ai.com/en/blog/wordpress-chatbot): 5-minute setup guide
- [Wix chatbot](https://bravos-ai.com/en/blog/wix-chatbot): an honest guide to AI chat on Wix. What Wix's built-in Smart Chat does (manual mode free; AI mode trained on site content plus hand-written rules) and how its AI-credit system really works, with official numbers: free plan 30 credits/day capped at 120/month, paid plans ~150-1,000 estimated actions per cycle, a pool shared with Wix's other AI tools, no guaranteed credit-to-conversation equivalence (Wix states it literally), and the AI chat switching to offline mode when credits run out. When the built-in chat is enough, when a specialized chatbot is needed (exact catalog filtering, unpublished documents, WhatsApp, answering in the visitor's language), a 4-question test to run on any vendor's trial, how to prepare a Wix site so any bot answers better, and the step-by-step install of an external widget (requires a paid Wix plan with a connected domain; Settings → Custom Code). Wix Stores catalog connects via CSV export or a synced Google Sheets document; a direct Wix Stores integration is on the Bravos roadmap
- [WhatsApp AI chatbot — how to build one without code](https://bravos-ai.com/en/blog/whatsapp-ai-chatbot): three real paths (n8n/Make/Zapier vs Meta Cloud API vs purpose-built platform), why ChatGPT alone is not enough for business, comparison vs Manychat/Wati/Respond.io, and the truth about Meta's July 2025 pricing (replies to customer-initiated chats are free, no monthly cap). Bravos AI plan from $23/month with direct Meta Tech Provider integration — no Twilio, no middleware fees
- [WhatsApp auto reply in 2026 — out-of-office, Business and AI](https://bravos-ai.com/en/blog/whatsapp-auto-reply): three real ways to set up an auto reply on WhatsApp — the static away message in the WhatsApp Business app, rule-based templates, and AI that understands the customer and answers with the business's own data. Includes the difference between automating personal WhatsApp (ban risk with unofficial apps like WhatsAuto) vs business WhatsApp (the official Meta path), setup steps, and Meta's July 2025 pricing change (replies to customer-initiated chats are free). Bravos AI is a verified Meta Tech Provider — no Twilio, no 60-day token refresh, no middleware fees
- [n8n, Make and Zapier for WhatsApp AI chatbots — the honest cost breakdown](https://bravos-ai.com/en/blog/whatsapp-chatbot-n8n-make-zapier): the 9 components you actually need to wire if you want to build a WhatsApp AI chatbot on n8n self-hosted, Make.com or Zapier (not the 2 the YouTube tutorial shows). Real monthly cost breakdown ($21-95/month depending on platform, vector DB and OpenAI usage), real developer hours (25-40h the first time), the Meta Cloud API gotchas tutorials skip (60-day token expiry, media_id download for inbound images/audio, 24-hour window, debugging without proper logs). When DIY actually makes sense (devs with spare time, custom ERP integration, micro volume) and when it does not (non-devs, agencies scaling to multiple clients, mission-critical 24/7). Comparison vs Bravos AI as a verified Meta Tech Provider with all 9 components managed on the platform side

## Technology

- **LLM**: OpenAI GPT-6 family (GPT-6 Luna on Starter, GPT-5.6 Luna on Pro, chosen per client on Enterprise; GPT-5.4-nano for filter extraction)
- **Embeddings**: text-embedding-3-small (1536 dims)
- **RAG**: PostgreSQL + pgvector, hybrid search (semantic + keyword + structured) with Reciprocal Rank Fusion
- **Reranker**: Voyage AI rerank-2.5
- **Multimedia**: AWS Bedrock (Claude) for image analysis, AWS Transcribe for audio, AWS Textract for PDFs
- **Backend**: FastAPI (Python), SQLAlchemy, Uvicorn
- **Frontend**: Next.js 15, React, Tailwind CSS, next-intl
- **Storage**: AWS S3
- **Payments**: Stripe
- **Email**: Resend
- **Hosting**: Hetzner

## Pricing

Bravos AI does not have a free tier. New customers start with a 7-day full PRO trial.

Prices are billed in the customer's currency: EUR, USD, BRL or GBP (Starter €19 / $23 / R$ 119 / £16 per month; Pro €49 / $59 / R$ 299 / £42 per month).

- **7-day PRO trial, no credit card required**: the full PRO plan for 7 days. It starts the first time the user opens the dashboard; no payment method is asked for. To keep going, the user adds a card (Starter or Pro) from the dashboard, during the trial or after it; the first charge happens when the trial days run out. If no card is added, the bot stops answering and nothing is charged; the account and its content are kept for 30 days after the trial ends and then deleted. One trial per email address. Fair-use cap during the trial: 500 bot replies per day per account.
- **Starter**: €19/month ($23 / R$ 119 / £16) — 1 chatbot on web and WhatsApp, unlimited messages, AI included, all integrations (PrestaShop, Shopify, Resales…), MCP connector to manage the bot from Claude/ChatGPT, image and audio in conversations, custom instructions, email support within 48h
- **Pro**: €49/month ($59 / R$ 299 / £42) — 3 chatbots on web and WhatsApp, unlimited messages across all bots, more advanced AI for more accurate replies, lead capture inside the chat, Human Takeover (a person steps into any conversation and replies from the panel, on web and WhatsApp), 100% customizable system prompt and tone, more capacity per bot (URLs, files, multimedia), priority support within 24h
- **Enterprise**: custom — tailored chatbots and limits, premium AI model of choice, guided setup with the Bravos team, guaranteed response time per contract

## Key URLs

- Website: https://bravos-ai.com
- App: https://app.bravos-ai.com
- API: https://api.bravos-ai.com
- Sitemap: https://bravos-ai.com/sitemap.xml

## Legal

- [Privacy Policy](https://bravos-ai.com/en/legal/privacy)
- [Terms of Service](https://bravos-ai.com/en/legal/terms)
- [DPA](https://bravos-ai.com/en/legal/dpa)