JevMade Sign in
← Back to guides

JevMade field notes / Official release announcement

Pi 1.0 adds small decision models to its coding workspace

Earendil's Pi 1.0 announcement includes Codemode support for non-chat models such as Jev alongside its existing coding models.

Original by EarendilIntegrations

Listen to this guide

JevMade’s plain-English explanation

0:00 /

AI narration

Credits

“Pi 1.0” by Earendil. Read the original source.

This expanded guide is an AI-narrated adaptation prepared by JevMade. It expands the source’s essential ideas, examples and caveats in JevMade’s own words and is not a word-for-word reading. The synthetic voice does not imitate the author or imply their endorsement.

Our summary

Pi is a coding-agent workspace designed to stay small while allowing extra features. Its official 1.0 announcement adds Codemode, where a program can combine calls to tools and models. This includes Jev, which answers narrow questions instead of writing prose, without replacing the larger model that plans and writes the code.

The release also lets extensions combine models behind one name, and changes how tools and conversations are managed. Its demonstration describes an extension that plans with Claude and implements with GPT, using Jev to decide when to switch. The linked code sends questions to Jev and passes its fixed-format answers back to the workspace.

These are release capabilities, not measured improvements in coding success or total cost. Pi Durable is introduced separately as an experimental package for longer-running applications. Connected providers still need credentials, and tools can have real effects. Adding a decision model does not make its answer correct or grant permission to act.

Key takeaways

  1. Use a decision model for a narrow question rather than treating it as a replacement for a coding model.
  2. Keep release capabilities separate from claims about accuracy, savings or better code.
  3. Trace each provider and tool involved; a local workspace does not make all model processing local.

The announcement's captions describe a Claude-to-GPT demonstration with Jev deciding when to switch. The archived router example differs: Jev chooses an initial OpenAI planning model, and a successful edit triggers the later switch. The integration code does not establish a reproduced run, measured savings or better code. Pi Durable is separately experimental.

Earendil · Original published

Read the original guide Opens the author’s site in a new tab.

Keep this for later

Sign in to bookmark experiments, guides and videos, and keep notes only you can see.

Continue to sign in

We’ll bring you back to this listing.