Shannon: One Model, Every Provider
A large merged change simplifies how Shannon talks to AI providers—one model setting instead of tiers—while adding structured SARIF output for exploit runs, followed by a string of README cleanups to keep docs aligned with the new setup.
Duration: PT2M19S
Episode overview
This episode is a short developer briefing from Shannon.
It explains recent repository work in plain language.
- Show: Shannon
- Published: 2026-07-31T06:02:53Z
- Audio duration: PT2M19S
Transcript excerpt
This excerpt keeps the crawler page concise. Listen to the episode or use the RSS feed for the full update.
Good day, and welcome to Shannon. It's July 31st, 2026.
The headline today is consolidation. The team collapsed a scattered, tiered model configuration into a single setting called Shannon AI Model, working across five providers, and used that cleanup to retire Google and Gemini support entirely, along with a couple of suggested models that didn't earn their keep.
That's the core of PR 402, a broad merge from ezl-keygraph touching setup, the CLI wizard, and the worker. Alongside the provider simplification, it adds real reliability work: in-session retry with backoff handled by the pi layer instead of custom Anthropic-specific logic, live credential validation for Bedrock…
The other major piece in that same PR is SARIF output. Exploit-mode runs now emit a proper SARIF 2.1.0 log, with structured code locations carried from the vulnerability queue straight into the report instead of asking agents to reproduce them. That's a meaningful reliability upgrade—locations and confidence scores…
The second theme is documentation catching up to that change. PR 403 synced provider options across the bug report template, the README, and the environment example. Then three separate README updates…
Wh…