Interference is a low-latency, unified AI inference platform designed for modern autonomous software engineering agents, developer tools, and high-throughput applications.
Instead of maintaining multiple vendor SDKs, juggling separate API keys, and dealing with unpredictable credit policies or currency exchange markups, Interference provides a single OpenAI-compatible drop-in gateway for leading foundation models—including GLM-5.3, Kimi K2.7 Code, Qwen 3.6 Plus / Coder, DeepSeek V4, and Grok 4.6.
Drop-in OpenAI Wire Compatibility: Standard POST /v1/chat/completions wire format supporting SSE token streaming, multi-step tool/function calling, structured JSON output, and reasoning tokens.
Harness Ready: Native, first-class configuration support for autonomous coding harnesses including OpenCode, Hermes Agent, Cursor, Aider, and Continue.dev.
Public vs. Authenticated Gateways: Fully transparent, public model catalog discovery (GET /v1/models) paired with secured, low-latency authenticated endpoints (POST /v1/chat/completions, POST /v1/embeddings).
Atomic Two-Phase Reservations: Every inference call reserves a temporary credit hold before upstream streaming begins and settles for exact consumed tokens upon completion. Disconnects or upstream timeouts never result in negative balance drift or lost funds.
Dual-Currency USD Credits: Unified USD credits ledger with zero-forex UPI payments (PhonePe, Google Pay, Paytm, BHIM) for Indian developers and international cards via Dodo Payments globally.