Pi v0.78.1: Ant Ling and NVIDIA NIM providers, MiniMax-M3 support, extension context improvements
Pi adds Ant Ling and NVIDIA NIM provider setup with MiniMax-M3 model support, richer extension context (mode detection and system prompt inspection), and a wide batch of fixes across provider compatibility, XSS sanitisation in HTML exports, SDK embedding, HTTP timeout handling, and TUI overlay rendering.
PUBLISHED2026-06-04
OBSERVED2026-08-21
AGE2mo
SOURCES1
- New providers: Ant Ling and NVIDIA NIM provider setup with documentation; MiniMax-M3 model support for the
minimaxandminimax-cndirect providers - Extension context:
ctx.modeexposes TUI/RPC/JSON/print mode;ctx.getSystemPromptOptions()lets extensions inspect base system prompt inputs - Containerisation: added containerization docs and Gondolin extension example for routing built-in tools into a local micro-VM
- Security fix: stored XSS in HTML session exports sanitised via scheme allow-list on Markdown link and image URLs
- SDK embedding fix:
createAgentSession()no longer fails withENOENTwhenpackage.jsonis absent next to the bundle entrypoint - HTTP timeout fix:
httpIdleTimeoutMsnow applies to all providers, not just OpenAI Codex; disabling timeout sends max int32 instead of 0 (which SDKs treated as immediate timeout) - Provider compatibility fixes: Amazon Bedrock blank text handling; Claude Opus 4.7+ deprecated temperature suppression; OpenAI GPT-5.5 unsupported minimal thinking; OpenRouter Kimi K2.6 thinking replay and developer-role handling; NVIDIA Qwen 3.5 122B NIM model restored
- TUI fixes: overlay focus restoration, tab width accounting in column slicing, footer branch display refresh in WSL, blank chat row from
renderShell: "self"with no component lines - Session handling: large JSONL session files read line-by-line instead of materialising the full file
COMMUNITY
No curated reactions recorded for this event. Facts and takes are kept in separate layers — community context is added by hand, never blended into the record above.