Shannon 2.0 now runs on the Pi harness

[ezl-keygraph]

Shannon 2.0.0 is now available on latest.

We've rewritten the engine behind Shannon's agents to run on the Pi harness (earendil-works/pi), an open-source agent runtime, replacing the Claude Agent SDK. That moves us off a closed-source dependency and gives us a foundation for supporting more models over time. For now, Anthropic's models remain what Shannon is built and tested against. After several weeks in beta, the migration is stable.

npx @keygraph/shannon@latest setup

npx @keygraph/shannon@latest start \
  -u <https://your-app.com> \
  -r /path/to/your/repo

What changed?

This changes how Shannon's agents run, not what Shannon does. For most users the workflow is unchanged:

• The pentesting pipeline and its phases are the same.
• Shannon still performs source-aware reconnaissance, vulnerability analysis, and real exploitation.
• Reports and workspaces stay in the same format.
• The existing setup and start commands work as before.

The only thing to check is the breaking changes below — they won't affect most setups.

Breaking changes

Because this is a major release, there are two migration notes to be aware of.

Google Vertex AI support has been removed

The Pi harness does not support Anthropic models through Vertex AI. Vertex users will need to move to Anthropic, AWS Bedrock, or an Anthropic-compatible custom endpoint.

The following environment variables are no longer supported:

CLAUDE_CODE_USE_VERTEX
ANTHROPIC_VERTEX_PROJECT
CLOUD_ML_REGION
GOOGLE_APPLICATION_CREDENTIALS

Run the setup wizard to configure a supported provider:

npx @keygraph/shannon@latest setup

The manual maximum-output setting has been removed

The following options are no longer supported:

CLAUDE_CODE_MAX_OUTPUT_TOKENS

Pi now manages model output limits directly, so this variable can be removed from your environment.

Try Shannon 2.0

If you try Shannon 2.0, let us know how it performs, especially if you notice changes in agent behavior, runtime, cost, authentication, or tool execution.

Please report problems through GitHub Issues.