Tkngate, a new open-source proxy tool, sits between AI agents and API providers to automatically manage authentication keys, spending limits, security threats, and service failover.

What happened: Tkngate is a reverse proxy that routes requests from autonomous AI agents through multiple providers (OpenAI, Anthropic, DeepSeek, Kimi, Groq) via a single endpoint. When one provider experiences downtime, it automatically switches to another. The tool also blocks prompt injection attacks, redacts sensitive information, compresses code in prompts to reduce token usage by up to 40%, and caches identical requests to avoid redundant API calls.

Why it matters: Autonomous AI agents today face three concrete problems: service outages that crash workflows, runaway token consumption that drains budgets (the example given is a rogue loop burning $200 in tokens), and exposed API keys stored in plain text. Tkngate addresses these by enforcing hard spending caps per session, masking physical API keys behind isolated virtual keys for teams, and applying rate limiting to prevent burst loops — all transparently between the agent and the provider.

What to watch: The tool includes a P2P token mesh feature that lets users donate spare API keys to a shared pool and gain priority access during outages, protected by AES-256 zero-knowledge encryption. It also offers a shadow mode that silently mirrors a fraction of production traffic to a different provider (for example, DeepSeek) to evaluate cost savings with zero latency impact on the primary request. The code is available on GitHub under the Apache 2.0 license.