pi · Issues· 233 open

`PI_OFFLINE` silently disables all provider model discovery — undocumented behavior contradicting its documented scope

openai-codex SSE parser buffers the whole response in one string, fatal heap OOM

Empty-body 400 from OpenAI-compatible gateways (opencode-go/deepseek-v4-flash) misclassified as context overflow → destructive auto-compaction destroys up to ~400k tokens

After compaction, stale signed thinking blocks are replayed every turn; Anthropic drops them (prefix_binding_mismatch) on every request

Emit session_compact_end event to extensions after clearing compaction state

regression: clipboard copy doesn't work anymore

Compaction hits summary output cap with GPT-6 Astra at max reasoning

Compaction is refused by Anthropic Claude Fable because of transcribed thinking blocks

Agent loop hangs forever when a provider stream stalls mid-response

Overflow image recovery uses a fixed 16 MiB budget that exceeds smaller provider request limits

mistral: add zai-glm-5-3, zai-glm-5 and zai-glm-latest to the catalog

mistral-conversations: empty content deltas open text blocks (GLM 5.x trailing whitespace)

Jumpy "jump to latest message" overlay

/hotkeys: ctrl+x (app.message.copy) description omits selection and /tree fallbacks

[openrouter] meta/muse-spark-1.3-contributor with thinking=max → 400 invalid_request_error from Meta (works on high)