Built by an independent developer, not a faceless tech giant
The story behind Saathi Bot: why existing AI chatbot tools were bloated, overpriced, and poor at Indian languages—and how we set out to build something better.
Why Saathi Bot Exists
In late 2024 and 2025, the market was flooded with hundreds of “Custom GPT” and “Chatbase-like” wrappers. Most of them shared the exact same flaws:
- Exorbitant 10x–20x Token Markups: Charging $40 to $100 per month for basic Q&A that consumed less than $0.50 of actual underlying LLM API tokens.
- Embarrassing Hallucinations: Answering questions about refund policies or pricing with fabricated guesswork instead of admitting when official docs had no answer.
- Terrible Indian Language Support: Stumbling, translating robotically, or completely hallucinating when real customers in India typed queries in natural everyday Hinglish (Hindi in Latin script) or conversational Hindi.
- Opaque Vendor Lock-in: Hiding prompts, blocking developers from inspecting conversation vectors, and preventing owners from bringing their own API keys.
I built Saathi Bot (“Saathi” means companion or reliable partner in Hindi) to fix these fundamental problems from first principles. Not with venture capital hype or fake team headshots, but with disciplined engineering:
A hard confidence gate (< 0.55 similarity rejection) that explicitly refuses to make up facts when documentation lacks an answer.
Cross-lingual embeddings and prompts that understand “refund aane me kitna time lagta hai?” as accurately as formal English.
Fair, transparent plans starting at ₹0 / ₹999 INR per month, plus Bring-Your-Own-Key support to eliminate SaaS markups completely.
Built on Next.js Turbopack, Supabase pgvector, Google Gemini embeddings, and Groq ultra-fast LPU inference (300+ tok/s).
Tushar Singh
Creator & Maintainer · Bengaluru, India
“When you rely on software to talk to your customers, you shouldn't have to wonder whether the answers are truthful or whether your credentials are safe. I built Saathi Bot to be the AI tool I personally wanted to embed on my own projects: fast, honest, and completely grounded in verifiable documentation.”