The Cloudflare for Autonomous AI Agents
Install • The Mesh • Features • Configuration • Docs
Tkngate is an enterprise-grade, zero-knowledge reverse proxy daemon that protects your autonomous agents and LLM budgets. It provides Strict Rate Limiting, Semantic Caching, and Universal Failover out of the box.
But its killer feature is the Deficit Round Robin (DRR) Token Mesh: BitTorrent for LLM Tokens.
If you run parallelized agents, you will quickly hit OpenAI's rate limits (429 Too Many Requests). Tkngate solves this by allowing developers and enterprises to safely pool their unused API keys into a decentralized, zero-knowledge mesh, instantly multiplying bandwidth for everyone.
The problem with sharing API keys in a pool is trust. How do you know a rogue node won't use your clean OpenAI key to process ToS-violating prompts, getting your account banned?
Tkngate solves this using an Economic Game Theory Ledger:
- AES-256 Zero-Knowledge Payload: When you donate an API key, it is encrypted locally. You never see the plaintext prompts passing through your node.
- Pre-flight Moderation WAF: Every prompt is scanned locally before being routed, protecting against prompt injections and PII leakage.
- The Slash: If a malicious node somehow bypasses the WAF and OpenAI flags the request, the HTTP response acts as a Fraud Proof. The malicious sender's public key is permanently blacklisted globally, and their stake in the mesh is slashed to zero.
Tkngate is a single binary with zero external dependencies (no Redis or Postgres required out of the box). Version 1.7.0 introduces a robust, interactive CLI loop for managing your mesh and budgets seamlessly.
git clone https://github.com/tkngate/tkngate.git
cd tkngate
# 1. Build the binary
go build -o tkngate
# 2. Setup your config
cp tkngate.example.yaml tkngate.yaml
# 3. Start the interactive CLI
./tkngateThe CLI will guide you through generating a secure Master Key (AES-256) if you don't have one, and present a professional interactive menu to start the proxy, manage budgets, or view mesh telemetry.
- Virtual Budgets (Virtual Keys): Issue
tkngate-sk-...keys to your agents with hard USD caps. If a loop goes rogue, the proxy instantly blocks traffic, saving your credit card. - Semantic Caching: Identical prompts are served from memory for
$0.00with zero latency. - Universal Fallback: If OpenAI returns a
500or503, Tkngate automatically transparently falls back to Anthropic or DeepSeek. - Context Compressor: Automatically strips comments and whitespace from code-heavy prompts, reducing token bills by up to 40%.
- Shadow Mode: Silently mirror 10% of your production traffic to a cheaper provider (like DeepSeek) to evaluate responses without impacting your main flow latency.
- Secure Telemetry API: Monitor all nodes and sessions locally with Bearer-token authenticated metrics and CORS lockdown.
| Topic | Link |
|---|---|
| CLI Reference & Architecture | docs/cli-reference.md |
| Budget System & Traffic Lights | docs/budgeting.md |
| DRR Token Mesh & Reputation | docs/drr-mesh.md |
| Enterprise Virtual Keys | docs/virtual-keys.md |
| Strict Rate Limiting | docs/rate-limiting.md |
Apache 2.0