Shannon Now Runs on the Pi Harness (Beta) #358
ezl-keygraph
announced in
Announcements
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
-
A few weeks ago in #348 we said the big June effort was rewriting Shannon's agent runtime onto the Pi harness and moving off
claude-agent-sdk. That work is now done and available for beta testing.You can run it today:
(or check out the feat/pi-harness-migration branch for a source build).
This is a beta. The pipeline, phases, and reports are unchanged. What changed is the engine underneath them. We'd love feedback before it lands on
latest.Why we did this
The number one request from this community has been multi-model support. Earlier workarounds (
claude-code-router, then a custom base URL override) helped, but they were patches on top of a runtime (@anthropic-ai/claude-agent-sdk) that was fundamentally built around a single provider.Pi (
@earendil-works/pi-coding-agent) gives us a real foundation for this. Provider and model now resolve through pi-ai's model registry, with the same three-tier override contract working across Anthropic, AWS Bedrock, and any Anthropic-compatible base URL:For now, only Anthropic (Claude) models are supported. The Pi harness gives us the foundation to go broader, and we're actively testing other models. More on that soon.
Breaking changes / migration notes
npx @keygraph/shannon@beta setupto configure one of these alternatives.CLAUDE_CODE_MAX_OUTPUT_TOKENSis gone. Remove it from your.env/config; pi manages output limits.What we want from you
npx @keygraph/shannon@betaand tell us if results, cost, or runtime feel different fromlatest.More multi-model news to come on this foundation.
Beta Was this translation helpful? Give feedback.
All reactions