One POST /v1/threads/:id/messages and you have persisted conversations, 100+ vendor connectors, semantic search, per-user spend caps, and tool dispatch.See the master pitch on the landing page →
Send the next user turn. qlaud loads prior messages into the model's context, persists the assistant reply, returns the response. No `messages` table to maintain. No retention cron. Pagination + GDPR-delete are API calls.
POST /v1/threads/:id/messagesLinear, GitHub, ClickUp, Notion, Stripe, Sentry, HubSpot, Salesforce, Zapier, 97 more — auto-enabled per end-user. They authorize their own account via a one-time hosted URL. Zero per-vendor wiring on your side.
/connectors · 105 liveVectorize-backed semantic search across every conversation in your account, scoped by end-user. No embedding pipeline, no pgvector index, no separate vector store bill.
GET /v1/search?q=…Mint a key per end-user with a hard `max_spend_usd` cap. The gateway enforces it BEFORE forwarding to the provider — over-cap requests return 402, no upstream burn. Pull per-user usage rollups for invoicing.
POST /v1/keysThe default AI-app stack is the same diagram in every Notion doc. qlaud collapses three of these four into managed primitives.
| Layer | Without qlaud | With qlaud |
|---|---|---|
| Auth | Clerk / Auth0 / Supabase Auth | Bring your own — qlaud doesn't replace this |
| Per-user billing | Stripe + custom proxy that meters tokens, enforces caps, emits invoices | POST /v1/keys with max_spend_usd; gateway enforces; GET /v1/usage for invoicing |
| Conversation state | Postgres `messages` table + index + retention cron + GDPR cascade-delete | Threads API persists each turn; pagination is a query param |
| Semantic search | Pinecone / pgvector + embedding pipeline that fires on every message | GET /v1/search?q=… — auto-indexed in Vectorize |
| LLM proxy | Express / FastAPI route that streams, retries, falls back across providers | qlaud AI Gateway under the hood; native shapes preserved |
| Tool dispatch | State machine that parses tool_use, calls the function, appends result, re-invokes | Built-in dispatch loop; meta-tools for dynamic discovery |
| Connectors | Per-vendor SDK + per-user OAuth + credential vault + per-tool dispatch path | 105 catalog vendors auto-enabled; hosted connect URL flow |
You spend a quarter writing this and you're still patching it next year. Or you change a base URL and add four lines on signup.
No tools array, no Linear setup, no DB write, no embedding job.
POST https://api.qlaud.ai/v1/threads/:thread_id/messages
Authorization: Bearer qlk_live_<your_end_user's_key>
{
"model": "claude-sonnet-4-6",
"max_tokens": 1024,
"content": "Create a Linear ticket: 'Login button broken on Safari iOS'",
"stream": true
}
# What happens, autonomously, inside one HTTP call:
# 1. qlaud loads prior thread messages into context
# 2. tools_mode defaults to "dynamic" (no tools array passed)
# → 4 meta-tools injected (qlaud_search_tools, get_tool_schemas,
# multi_execute, manage_connections)
# 3. Model calls qlaud_search_tools({intent: "linear ticket"})
# 4. Catalog returns Linear (auto-enabled, not yet connected for this user)
# 5. Model calls qlaud_manage_connections({action: "connect", tool: "qlaud-mcp/linear"})
# 6. qlaud mints a hosted URL, model relays it: "open this to authorize Linear"
# 7. End-user pastes their Linear API key in the qlaud-hosted form
# 8. Model calls linear/create_issue → ticket lands in Linear
# 9. qlaud persists the assistant reply, indexes it for search,
# debits the user's wallet, returns the streamed responseqlaud exposes two API surfaces. Same auth, same wallet, different power-vs-simplicity tradeoff.
Pure Anthropic-shape passthrough. Whatever tools array you send is forwarded verbatim. Whatever you don't, the model has no tools.
Right when: your existing app already manages history, tool dispatch, and any third-party integrations.
Send the next user turn. qlaud handles the rest. Default tools_mode is dynamic — the model auto-discovers + invokes any of 105 catalog tools.
Right when: you're building a chatbot, an agent, or any AI UX with end-user identity.
The data plane is Cloudflare's — the same primitives Cloudflare uses for their own production services. Globally edge-distributed, atomic per-customer wallet, native vector search. Your data is exportable via API any time. Security details.
Top up $5, mint a master key, ship a chatbot in 200 lines.