Every model.
Private by default.
An OpenAI-compatible API for coding agents. Every model runs with zero data retention — prompts are never stored or trained on. t2000/auto picks the cheapest capable model per call, so most of your bill runs at open-model prices.
Automatic model routing
cuts your bill.
Set model: t2000/auto. Routine steps run on a cheap open model; hard steps escalate to a frontier one. You pay the price of the model that served.
Runs bulk steps on zai/glm-5.2; escalates long context, planning, and retries-after-failure to anthropic/claude-sonnet-5.
The same router, open models only — hard steps go to moonshotai/kimi-k2.7-code. Your prompts never reach a closed-model provider.
Every response names the model that served it and why: x-t2000-served-model + x-t2000-route-reason. You can audit every charge.
moonshotai/kimi-k2.7-code — works in any OpenAI-compatible tool.npm i -g @t2000/code. Router by default, per-repo privacy pinning, wallet tools in-session.Every model.
Two privacy tiers.
Live catalog — USD per 1M tokens, straight from GET /v1/models. Public, no key.
Every model, zero data retention: providers are contractually bound not to store or train on your prompts.
Inference runs inside a hardware enclave (GPU-TEE); every response carries a signed receipt anchored on Sui — you can prove where it ran.
Or pay per call
over x402.
Agents can pay per call from their wallet’s USDC — no key, no account, gasless. The gateway handles 402 → pay → retry automatically.
$ t2 pay https://x402.t2000.ai/t2000/v1/chat/completions \ --data '{"model":"zai/glm-5.2","messages":[…]}' \ --max-price 0.10
Every confidential response
comes with proof.
The proof is anchored on Sui and read from a public fullnode — we can’t forge it, and you don’t have to trust our server.
Private by default
Every model is zero data retention — prompts and outputs are never stored, logged, or trained on.
Confidential tier
phala/* models run in a verified GPU-TEE. The gateway attests the upstream before forwarding — and fails closed if it can't.
Signed, Sui-anchored receipts
Every confidential response commits its hash on-chain — tamper-evident and publicly timestamped.
Verify it yourself
t2 verify checks the Sui anchor, receipt signature, and TDX quote on your machine — no trust in our server required.
Point any OpenAI client at it.
Two environment variables. Most tools repoint with zero code changes.
export OPENAI_BASE_URL="https://api.t2000.ai/v1" export OPENAI_API_KEY="sk-…"
One agent stack.
Every model, private.
One base URL.
Keys are free — sign in and create one. Free daily coding on kimi-k2.7-code before you spend a cent.