Core
Text chat (RAG)
Question → embedding → vector search filtered by project_id → top-k chunks → LLM → answer with sources. One endpoint: POST /v1/chat.
pyetja→embedding→match_chunks→llm→answer + sources
Launch AI support in minutes.
Your docs become
the API.
Upload a PDF or URL, get a pk_xxx key, and your chatbot is live on your site in under 5 minutes.
sdk · 1keyai.orgprivate beta
// playground
A live tour of the dashboard. Switch tabs to walk through the pages.
1KeyAI
dashboard.1keyai.orgB
Built on the best
OpenAI
Anthropic
Supabase
Firecrawl
ElevenLabs
Vercel
FastAPI
Scalar
OpenAI
Anthropic
Supabase
Firecrawl
ElevenLabs
Vercel
FastAPI
Scalar
What is 1KeyAIThe full RAG stack,
The full RAG stack,
hidden behind one key.
RAG from PDF + URLText + voice chatClaude · GPT · GeminiRow Level SecuritySelf-hostableOpen API (OpenAPI)
If you wanted to build an AI chatbot today that answers from your documents, you'd wire up OpenAI for the LLM, Pinecone for vectors, ElevenLabs for voice, and Firecrawl for websites. Weeks of work. Five keys to babysit.
1KeyAI collapses that into one endpoint. Create a project, upload a PDF or URL, get a pk_xxx key, and POST to /v1/chat from your site. Behind the key: embeddings (text-embedding-3-small), pgvector isolated with Row Level Security, and an LLM router across Claude, GPT, and Gemini.
Goal: from signup to a working chatbot in under 5 minutes. API-first — we don't impose a UI or ship a widget. You decide what the visitor sees; we keep the brain.
What you get behind one key
One endpoint loads your docs. Another returns answers with sources. Five AI services run on our side, not yours.
5 services · 1 key
LLM
Claude · GPT · Gemini
Vectors
Supabase pgvector
Embeddings
OpenAI 1536
Voice
ElevenLabs STT+TTS
Ingest
PDF · Firecrawl URL
How it works
From signup to a working chatbot — in under 5 minutes.
- 01
Create a project
Sign up and create a project — get your pk_xxx key instantly, with a one-click copy.
- 02
Load knowledge
Upload a PDF or paste a URL. We extract text, chunk it, generate embeddings (1536) and store in Supabase.
- 03
Test in playground
Chat with your bot right inside the dashboard. See the sources behind every answer.
- 04
Copy the snippet
Paste the code into your site. One POST /v1/chat. The bot works — text or voice.
What's inside
Everything behind one API key.
You don't touch embeddings, vector DBs, STT/TTS, or LLM routing. We hide them all behind a single endpoint.
Voice
Real-time voice chat
ElevenLabs hears and speaks. Our backend keeps the brain — RAG + LLM. Same docs, no double setup.
02→
Image
Image generation
POST /v1/image with a prompt. Router across OpenAI, Flux, and Stability — the keys live on our side.
03→
Ingest
PDF & URL, same shape
Upload a PDF (pypdf) or paste a URL (Firecrawl). Same embeddings, same hierarchy: documents → chunks.
04→
LLM Router
Claude, GPT, or Gemini
One dropdown. Same docs. Default: claude-sonnet. Switch with one request parameter.
05→
VideoRoadmap
Async video
Google Veo / Runway / Sora after the hackathon — async with job_id + polling/webhook. Same pattern as image.
06→
Integration
One endpoint. Nothing else.
FastAPI generates OpenAPI automatically. Scalar turns it into beautiful docs with copy-ready code samples — your key pre-filled. Coming: Fern-generated SDKs for JS, Python, and PHP.
const r = await fetch("https://api.1keyai.org/v1/chat", {
method: "POST",
headers: {
Authorization: "Bearer pk_live_••••",
"Content-Type": "application/json",
},
body: JSON.stringify({ message, llm: "claude" }),
});
const { answer, sources } = await r.json();Security
Your data, isolated from the ground up.
One database, two layers of defense. No developer can ever see another's data — not even by mistake.
Site / Chat widget
POST /v1/chatAuthorization: pk_xxxBackend
layer 01
Verify
pk_xxx→ resolve project_idlayer 02
match_chunks(project_id, q)— filter inside querypgvector + RLS
policy own chunks
policy own documents
policy own projects
DB-level enforcement
01
Filter by project_id
Vector search always filters inside the query — not after. No chunk from another project ever reaches the LLM.
02
Row Level Security
Supabase RLS enforces isolation at the database layer. Even if the code slips, the DB itself blocks cross-project leaks.
03
Server-side keys
OpenAI, Anthropic, Firecrawl, ElevenLabs — all live on the backend. The frontend never touches them.
match_chunks · inside the query
select c.id, c.content, d.source from chunks c join documents d on d.id = c.document_id where c.project_id = p_project_id -- isolation happens here order by c.embedding <=> query_embedding limit match_count;
Do you ship a ready-made chatbot UI?+
No, by design. We're API-first — we give you the power, not the look. Every site has its own style; developers build the UI to match. A lightweight SDK is coming, not a styled widget.
How is data isolated between projects?+
Two layers: vector search filters by project_id inside the query (not after), and Supabase RLS enforces isolation at the database layer. Even if the code slips, the DB blocks cross-project leaks.
Which LLMs are supported?+
Default: Claude (claude-sonnet). Dropdown also covers GPT, with Gemini optional. Switch with one request parameter — same docs, no reconfig.
How fast can I get started?+
Under 5 minutes from signup to a working chatbot. No mandatory config — defaults just work. The code snippet ships with your key pre-filled.
Can I use my own OpenAI/Anthropic key?+
Not in the MVP. After the hackathon: BYOK for enterprise, plus per-user API keys (every user gets their own key with the provider — no rate-limit bottleneck).
What about video?+
Roadmap after the hackathon. Providers slot into the router (Google Veo / Runway / Sora) with the same pattern as image. Async (job_id + polling/webhook) because it takes 30s–several minutes.
Private beta
Grab the key. Build the bot.
Free to start. No card. No setup. Just a key and your documents.
· Open source· Self-hostable· API-first