Your WhatsApp. Your AI keys. Zero markups.
Drop OpenAI, OpenRouter, or AnythingLLM nodes into any flow. Add web scraping and browser-agent capabilities. Pay your LLM provider directly — we don't take a cut.
BYOLLM architecture
Your API keys. Your provider contract. Your model choice. We never proxy, meter, or markup.
Three LLM nodes
OpenAI, OpenRouter, AnythingLLM — covering cloud, multi-provider routing, and self-hosted.
Web scraping built in
Jina Reader or CSS selector modes. Feed fresh web content into your conversations without a backend.
Browser agent
Playwright + LLM combined. Automate multi-step tasks: check order status, submit forms, extract data.
Why BYOLLM matters
AI markups add up fast. Platforms that bundle AI charge 2–3× what you'd pay OpenAI directly, because they need margin on top of their infrastructure. BYOLLM eliminates that tax — you get our orchestration layer at a flat platform fee, and pay OpenAI (or anyone) at wholesale rates.
- No per-token markup from WapiSnap
- Switch models without changing flows
- Use private deployments when compliance demands it
AI with guardrails
AI makes chatbots useful; guardrails keep them sane. Our flow engine wraps every LLM call with hard limits — deterministic step counts, repeat-visit caps, cooldowns, and automatic halt on human takeover.
Frequently asked questions
What does BYOLLM mean?+
Bring Your Own LLM. Plug in your own API key for OpenAI, OpenRouter, or any OpenAI-compatible endpoint (including self-hosted AnythingLLM). We don't proxy, meter, or mark up AI calls — you pay your provider directly.
Which providers are supported out of the box?+
OpenAI (gpt-4, gpt-4o, etc.), OpenRouter (for Claude, Gemini, Llama, Mistral and hundreds more), and AnythingLLM (for local / private deployments).
Can the AI actually browse the web?+
Yes. The Web Scrape node extracts content from any URL using Jina Reader or custom CSS selectors. The Browser Agent node combines Playwright with your LLM to complete interactive tasks — form fills, multi-step navigation, extractions.
What prevents the AI from running wild?+
Every flow run caps at 50 steps and limits repeat visits to the same node. AI nodes are just one step each — a runaway prompt loop can't spam your customer.
Do you store my AI prompts?+
Every flow run logs inputs and outputs for debugging (stored in your workspace only). Credentials are encrypted at rest with AES-256-GCM. If you need stricter retention policies, use a self-hosted LLM via AnythingLLM.
Your WhatsApp. Your AI. Your rules.
Join businesses automating WhatsApp without vendor lock-in.