Peer-to-Peer AI
Pistachio
Direct-to-provider AI inference.
No middleman. No broker tax.
Built for capex stability and the lowest cost AI inference available today.
Why We Built It
DRM3 is always looking for arbitrage on AI pricing.
We started with Venice DIEM when it was $80. Today it is $900. Price asymmetry in AI inference is real and constant. Morpheus is pound-for-pound the least expensive option right now. Roughly 30% less per DIEM equivalent than Venice today. But the Morpheus gateway routes through a third-party AI broker who takes a cut.
We built Pistachio to remove the broker entirely. True peer-to-peer: your machine connects directly to the AI provider. One less hop resolves latency and security concerns, and the broker tax disappears. The result: MOR staking for inference per DIEM costs roughly 60% less than Venice DIEM today.
Pistachio replaces the Morpheus consumer node. We rebuilt the Lumerin node for maximum savings and added user-friendly features: auto-reclaim and auto-staking to renew sessions without manual intervention. Then we added multi-tenant gateway key management to the inference API. One app replaces the consumer node and gives anyone the ability to be their own gateway provider, for themselves or others.
One command: pistachio serve
The Economics
MOR is collateral, not payment. Lock MOR to open an inference session. Your full stake returns when the session closes. Providers are paid by Morpheus protocol emissions, not from your deposit. Your only real cost is ETH gas (~$0.01 per cycle on Base).
Venice DIEM
Centralized, dollar-per-day pricing
baseline
Morpheus Gateway
Stake-based, but routed through a broker
~30% less
Pistachio (direct)
Peer-to-peer, no broker, no tax
~60% less
Morpheus DIEM equivalent is fluid (stake-based) compared to Venice's fixed dollar-per-day. Percentages reflect current pricing and may shift.
How It Works
Wallet = Identity
Every request is signed with your wallet's derived key using EIP-712. The network recovers your identity from the signature. No passwords, no tokens, no shared secrets. Cryptographic identity on every call.
NFT-Gated Access
Capacity is determined by Pistachio Pass NFTs on Base. Claim a free pass for 20 MOR. Upgrade to Bronze, Silver, Gold, or Diamond for higher limits. Larger stakes mean longer sessions, fewer transactions, lower gas.
Provenance on Every Response
Every inference result carries a signed receipt: the model version, the provider, the timestamp, the session. The same provenance protocol that powers the rest of DRM3. You can verify what model answered and who provided it.
OpenAI-Compatible API
Pistachio exposes a standard /v1/chat/completions endpoint. Any tool that works with OpenAI works with Pistachio. Point your existing code at localhost:19377 and go.
50+ Models
Every size class. Flash to frontier.
Small / Flash
LLaMA 3.2 3B, Qwen 3.5 9B, DeepSeek V4 Flash, GLM 4.7 Flash, Gemma 4 26B/31B. Fast inference for classification, extraction, tagging, and bulk work.
Mid / Large
LLaMA 3.3 70B, Mistral 3.1 24B, Qwen 3 Next 80B, GPT-OSS 120B, DeepSeek V4 Pro, Qwen 3 235B, Kimi K2.6, Hermes 3 405B. Quality reasoning and general-purpose.
Specialized
Qwen 3 Coder 480B (code generation), BGE-M3 (embeddings), TEE-secured models on Secret Network (hardware-verified inference). Most models have :web variants with search and tool capabilities.
New providers register models continuously. The marketplace is live and growing.
Server-Ready
Runs Headless on Any Linux Server
No GUI required. Run pistachio serve in a systemd unit or Docker container. It auto-restakes expiring sessions with zero manual intervention. Used in production on Fly.io and bare-metal Linux.
Parallel Request Capacity
Each staked session provides one concurrent request lane. Open multiple sessions for parallel inference. The built-in plan manager handles this automatically based on your staking template.
Minimal Capital Exposure
Sessions last minutes to hours, not days. MOR returns immediately on close. Use a dedicated low-balance hot wallet with only the MOR needed for staking. Price volatility risk is minimal at these durations.
Pistachio reads all data directly from Base-chain smart contracts via RPC. No intermediate cache, no API dependency, no third-party data layer. Ground truth is the blockchain, and Pistachio reads it in real time — sessions, marketplace bids, provider endpoints, balances, and NFT capacity.
For a standalone Morpheus block explorer, see Cashew.
Access Passes
Your NFT is your capacity tier.
Pistachio uses consumer passes (ERC-721 NFTs on Base) to determine how much inference capacity you get. Claim a free pass by staking 20 MOR. Upgrade to higher tiers for more capacity, longer sessions, and lower per-request gas costs. Passes are additive. Hold multiple to stack capacity.
Free
20 MOR
Bronze
+200 MOR
Silver
+1K MOR
Gold
+5K MOR
Diamond
+20K MOR
Get Started
Install Pistachio in one line.
Mac - Desktop app + CLI
$ brew tap drm3labs/drm3 && brew install --cask pistachio-desktop
Linux / WSL2
$ curl -fsSL https://drm3.network/install | sh
macOS (Apple Silicon + Intel), Linux, and Windows via WSL2.