One unified API
Talk to OpenAI, Anthropic, Gemini, Sarvam, ElevenLabs and more through a single, consistent endpoint. Switch providers without rewriting your code.
Open-source AI middleware
Leverage Kaapi to ship AI features. So that you can focus on impact, not plumbing.
Kaapi enables NGOs to harness LLMs through a unified platform — multi-provider, multimodal, and built around the workflows the social sector actually runs.
10+
AI providers supported
7+
Indian languages, with voice
100%
Open source
Build Responsible AI for the Social Sector with Kaapi
A unified middleware that gives programme leads and engineers the same building blocks.
Multimodal by default
Text, image, document, voice. Send a voice note in, get a voice note back. Score student submissions made of text plus prototype photos. The pipeline handles it.
Evaluations built in
Curate a golden dataset, run it across prompts and models side by side, and pick a configuration with evidence instead of gut feel.
Guardrails built in
PII masking, slur and ban lists, topic relevance, gender-bias removal, toxicity detection. Compose them per use case.
Indian languages, including voice
Hindi, Odia, Assamese, Gujarati, Telugu, and more — across text and speech-to-speech. Built specifically for audiences that don't read English.
Console + API
Programme leads use the console. Engineers use the API. Same backend, same AI configs. Move between them as your team grows.
What you can build with Kaapi
Three patterns the social sector ships every day — production-ready out of the box.
01
Knowledge-base chatbots
Anchor a chatbot on your own documents — government circulars, training material, FAQs — and ship it to WhatsApp through Glific or your own client. Curate a golden dataset, run evals, lock the configuration, go to production.
02
Speech-to-speech bots in local languages
For lower-literacy or younger audiences, voice is the primary modality. Kaapi supports speech-to-speech with knowledge-base lookup in major Indian languages — as a stitched pipeline you control, or a single chain endpoint with intermediate callbacks.
03
AI assessment pipelines
Score lakhs of student or beneficiary submissions a year with rubric-based LLM evaluation. Multimodal in (text plus images and PDFs), structured out (per-metric scores, reasoning, student-facing feedback). Compare up to four configurations on the same dataset.
Start building
Ship your first AI feature this week.
Console for programme leads. API for engineers. Same backend, same configs.
Our partners
TAP
Built for the social sector. Open source.
Kaapi is a Project Tech4Dev initiative.