Integrated Mantice runtime for Pi: routed models, mechanical request reduction, accountable goals, managed subagents, and an event-driven activity sidebar

Subagent delegation for pi: single, parallel, and chained runs in isolated pi processes, with a user-defined agent roster, model resolution, and slash commands.

Persistent responsive message history sidebar for Pi

Session recap panel for pi coding agent. Always-visible widget showing the current goal and last few conversation turns with live streaming updates. Auto-selects the fastest, cheapest summarization model using pi-bench benchmark data.

The complete Alibaba provider for pi — Plan subscription + Cloud pay-per-token, International + China endpoints, Anthropic + OpenAI shapes, native thinking levels. Qwen 3.8 Max, Qwen 3.7/3.6, DeepSeek V4 Pro, Kimi K3, GLM-5, MiniMax M3.

Non-blocking background watcher for pi: start a process, SSH poll, or log tail and get pinged in-session on milestones, failure, or exit. Claude Code's Monitor tool, with conditional delivery.

LLM benchmark toolkit for pi coding agent. Probes every available model with real streaming API calls and ranks by latency, cost, and output quality. Provides curated model chain and blacklist for smart model selection in pi-recap and other extensions.

Complete model providers for pi — Alibaba Model Studio (Plan subscription + Cloud pay-per-token, Anthropic + OpenAI + Responses API shapes, native thinking levels. Qwen 3.8/3.7/3.6, DeepSeek V4, Kimi K2.6/K2.7, GLM-5.1/5.2, MiniMax M2.5) and DeepSeek offi

Generate and edit video in pi using Alibaba video models: HappyHorse 1.1/1.0, Wan 3.0, Wan 2.7 (T2V, I2V, R2V, Video-Edit). Auto-selected defaults, per-task-type overrides.

Automatically updates a JSON file with the last modified dates of files in a Git repository.