Shannon: OpenAI Joins the Provider Lineup
A single broad pull request adds OpenAI as a first-class model provider in Shannon, introducing tiered GPT-5.6 models mapped to workflow complexity while keeping existing Anthropic, Bedrock, and custom endpoint paths intact.
Duration: PT2M5S
Episode overview
This episode is a short developer briefing from Shannon.
It explains recent repository work in plain language.
- Show: Shannon
- Published: 2026-07-20T06:03:38Z
- Audio duration: PT2M5S
Transcript excerpt
This excerpt keeps the crawler page concise. Listen to the episode or use the RSS feed for the full update.
Good morning. It's July 20th, 2026, and this is Shannon.
Today's story is one big architectural addition: PR 392 brings OpenAI in as a fully supported provider, sitting alongside Anthropic, custom endpoints, and AWS Bedrock rather than replacing any of them.
The key design choice here is tiered routing. Instead of running an entire scan on one model, the change maps Shannon's existing cost tiers — small, medium, and large — onto three specific GPT-5.6 variants: luna, terra, and sol. Each workflow phase, and each task spawned underneath it, now gets routed to the model…
Worth noting: Pi remains the internal agent and tool harness doing the orchestration. OpenAI is plugged in through Pi's native "OpenAI Responses" adapter, so this isn't a rip-and-replace of the agent layer — it's a new backend option feeding into the same harness.
The PR description also mentions deterministic completion gates and exploitation-related logic, though the feed cuts off before full detail. Given the breadth of areas touched — setup, packaging, documentation, and the core provider logic — this reads as a substantial, multi-surface change rather than a narrow patch.
What to remember: if you're touching…