Connect
A single secure entry point to AI providers, open‑source models, and MCP servers. Choose a provider, set policies, and integrate with Swfte Builder, your own agents, or any workflow in minutes.
OpenAI, Anthropic, Google, and open‑source models (Llama, Mistral, etc.) via a unified API.
Deploy MCP servers, manage tools, and expose secure capabilities to agents with auditing.
Smart routing, failover, cost controls, rate limits, and observability (logs, traces, metrics).
Run in Swfte Cloud or fully on‑prem with your own keys, VPCs, and private networking.
Simple by design
Call your configured model once—Connect handles routing, retries, and logging.
// Example: call the configured model once; Connect handles routing POST /api/connect/completions { "model": "auto", "input": "Summarize the customer email and propose a helpful reply." } // Response { "output": "Here is a concise summary...", "model": "gpt-4o-mini", "logs": { "provider": "OpenAI", "latencyMs": 352 } }
Unified model router
- One endpoint across providers with automatic health checks and regional routing.
- Smart fallbacks and retries on rate limits, timeouts, or capacity events.
- Policy controls for costs, models, guardrails, and tenant/role restrictions.
- Transparent usage with logs, traces, and per‑request metadata.
- Bring‑your‑own‑keys (BYOK) or centralized billing with budgets and alerts.
- Cloud, on‑prem, and hybrid with customer‑managed keys.
Connect from your tools
Expose Connect as an MCP server so your Cursor agents can call any configured model or tool with enterprise guardrails.
- Create a Swfte API token in your dashboard.
- Add the MCP server config to your Cursor settings.
- Restart Cursor and run a quick test prompt.
Call Connect from extensions, back‑end services, or internal apps using simple HTTPS.
Route prompts to the best available model with safe defaults and structured logging.
curl -X POST "/en/api/connect/completions" -H "Authorization: Bearer <SWFTE_API_KEY>" -H "Content-Type: application/json" -d '{ "model": "auto", "input": "Draft a polite follow up for this prospect about pricing." }'