-
Notifications
You must be signed in to change notification settings - Fork 2
Description
Description
When Pi runs with the default model from config (no /model set override), the footer only shows 🏷 dir: pi-test without any model name. When a model override IS active, the footer correctly shows the model (e.g. 🏷 dir: pi-test | openai-codex/gpt-5.4 or 🏷 dir: pi-test | sonnet 4.5).
Steps to Reproduce
- Ensure no model override is set (
/model clear) - Send a prompt to the Pi chat
- Check the footer on the response
Expected
Footer shows the model name from Pi's response metadata, e.g. 🏷 dir: pi-test | gpt-5.4
Actual
Footer only shows 🏷 dir: pi-test — no model name.
In contrast, when a model override is set (e.g. /model set openai-codex/gpt-5.4), the footer correctly shows 🏷 dir: pi-test | openai-codex/gpt-5.4.
Error responses also show the model: 🏷 dir: pi-test | sonnet 4.5.
Analysis
The PiRunner populates self.meta with model/provider info in the translate() method when handling AgentStart events. The issue may be that:
- The meta fields are only populated from
run_options.model(the override) andself.model(config), but whenself.modelisNone(no model in Pi config), the meta model field is empty - Pi CLI's response JSONL includes the actual model used in
message_endevents (e.g."model": "gpt-5.4"), but Untether may not be extracting it from there
Affected Files
src/untether/runners/pi.py— meta population intranslate()and/orStartedEventcreationsrc/untether/markdown.py— meta_line footer rendering
Testing Context
Found during v0.35.0rc12 Pi integration testing. The model name consistently appears in the footer when using /model set overrides but never appears when using the default config.