Jev directory
CommunityJev Agents & MCP
@pi-vault/pi-providers
Pi extension that registers MiniMax, StepFun, and Command Code as custom providers and adds a typesafe_decide tool answering typed Noul, Choice, and Score questions through TypeSafe Jev, with a Command Code fallback.
- Reading time
- 2 min
- Published by
- Community
- Author
- pi-vault
- Last verified
- Oct 2026
Used inJev for Software & DevTools
- Languages
- TypeScript
- Platforms
- CLI
- Capabilities
- Open sourceRequires API key
Highlights
- Adds minimax-openai, minimax-openai-cn, stepfun-ai, and command-code providers, each with its own API base.
- MiniMax-M3 and step-3.7-flash accept image input; MiniMax-M3 carries a 1,000,000-token context at $0.60/$2.40 per 1M tokens.
- typesafe_decide answers Noul, Choice (2-255 options), and Score (2-10 levels) questions with structured probabilities, not text.
- Calls TypeSafe System One directly with TYPESAFE_API_KEY, or makes one fallback request to typesafe/jev through command-code.
- Ships an 86-model Command Code snapshot captured 2026-09-30; output is capped at min(32,768, context window).
Watch out
Needs Node 24.15 or newer, and Command Code support needs Pi 0.86+. Model pricing is a dated snapshot that can drift, direct 400 or 422 responses never fall back, and the tool offers no enforced-ZDR mode because typesafe/jev has no ZDR-capable Command Code upstream.
Questions about this entry
Short answers drawn from the linked project or page.
MiniMax M3 and M2.7 variants under both the global minimax-openai and China minimax-openai-cn endpoints, StepFun Flash models as stepfun-ai, and Command Code models as command-code.
Run pi install npm:@pi-vault/pi-providers, then /reload. After reloading, select the provider in Pi's model picker; there is no provider-specific prompt syntax, and image input works with MiniMax-M3 and step-3.7-flash.
A Pi tool, not a chat model, so it never appears in the model picker and never accepts a chat prompt. It takes a JSON state plus one or more named questions and returns structured probabilities: Noul (yes/no with optional criteria), Choice (2-255 options), and Score (2-10 ordered levels).
It calls TypeSafe System One directly when TYPESAFE_API_KEY or a host-resolved typesafe credential is present; otherwise it makes a single fallback request to typesafe/jev through command-code. Direct 400 and 422 responses and cancellation are terminal and never fall back.
MINIMAX_API_KEY, MINIMAX_CN_API_KEY, STEP_API_KEY, CMD_API_KEY (with COMMAND_CODE_API_KEY as a secondary alias), optional TYPESAFE_API_KEY, optional CMD_ZDR for Command Code chat requests only, and CMD_MODELS_URL, which changes only where the live model catalog is fetched.
No. System One requests never inherit CMD_ZDR because typesafe/jev has no ZDR-capable Command Code upstream, so the tool provides no enforced-ZDR mode. CMD_ZDR applies only to Command Code chat-provider requests.
MiniMax-M3: 1,000,000-token context, 512,000 max output, $0.60/$2.40 per 1M tokens. MiniMax-M2.7: 204,800 context at $0.30/$1.20. step-3.7-flash and both step-3.5-flash variants: 256,000 context, with step-3.7-flash at $0.20/$1.15 and step-3.5-flash at $0.10/$0.30. Command Code models vary.
From the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.