Services · private-inference

Maple AI

End-to-end-encrypted AI inference an agent can drive through an OpenAI-compatible API — TEE-attested privacy, Bitcoin-accepted subscriptions.

What it is

Maple AI (by OpenSecret) is private inference: every prompt and response runs inside a Trusted Execution Environment with end-to-end encryption, so the content is unreadable to anyone — including Maple. The server code is open source and the TEE attestation is cryptographically verifiable (“don’t trust, verify” is their own framing). For an agent, the entry point is Maple Proxy: a lightweight, self-hosted proxy (Docker image, or bundled in the desktop app) exposing standard OpenAI-compatible endpoints (/v1/models, /v1/chat/completions) against the encrypted backend — so any OpenAI client library, LangChain, or LlamaIndex works unchanged.

It is the privacy-axis peer of PPQ.AI and Routstr: where those are sats-native and pay-per-call, Maple is subscription-based but provably blind to your content.

When to use it

• An agent’s workload is content-sensitive — prompts that shouldn’t be readable by any provider, with hardware-attested proof rather than a policy promise.
• The operator wants mainstream tooling compatibility: one API key, OpenAI-compatible calls, drop-in with existing frameworks.
• Models on offer are open-weight families (seven at last check, including kimi-k2-6, gpt-oss-120b, deepseek-v4-pro, llama3-3-70b) — pick Maple for privacy, not for closed frontier models.

Dependencies & payment

Dependencies: a Maple account (email / Google / GitHub — no KYC) on a Pro, Team, or Max plan (API access starts at $20/month, with usage credits on top), an API key from the account’s API Management page, and the Maple Proxy running locally or in infrastructure (Docker: ghcr.io/opensecretcloud/maple-proxy, pointed at enclave.trymaple.ai). Payment: card, or Bitcoin at a 10% discount — yearly subscriptions only, via the human checkout. The agent drives the API; the operator funds the account — there is no sats-per-call path here.

Quick start

Subscribe (Pro+), create an API key, run the proxy (docker run -p 8080:8080 -e MAPLE_BACKEND_URL=https://enclave.trymaple.ai ghcr.io/opensecretcloud/maple-proxy:latest), then point any OpenAI client at http://localhost:8080/v1 with the key as the Bearer token.

Gotchas

• The Bitcoin leg is not agent-automatable: Bitcoin pays the subscription (yearly plans, dashboard checkout), not the API calls. An agent can’t top up Maple in sats mid-run — for sats-native pay-per-call inference, see Routstr or PPQ.AI.
• Streaming-only API — the proxy currently supports streaming responses only; non-streaming clients need adjustment.
• Subscription + credits is a commitment model, not pay-per-use — wrong shape for sporadic workloads.
• Privacy claim is strong but scoped: the TEE protects content from the provider; the account layer (login identity, billing) is ordinary account data.