Pi 1.0 and Pi Durable released with TypeScript port and durable execution features
Pi 1.0 expands the minimalist agent harness with MCP support, deferred tool loading, and full-screen mode, while Pi Durable ports the system to TypeScript with checkpointing, multi-conversation concurrency, and hot-swapping.
1 source · cross-referenced
- Pi 1.0 adds native MCP and Jev support, deferred tool loading, and cache warming for Anthropic models.
- Pi Durable ports Pi to TypeScript, enabling crash survival via checkpointing and stateful component externalization.
- Pi Durable supports pluggable storage backends, concurrent parallel conversations, and dynamic tool/extension hot-swapping.
- Pi Durable’s state sync allows multiple users or UIs to connect to the same agent simultaneously.
Pi 1.0 introduces native support for the Model Context Protocol (MCP) and Jev, alongside extension support for virtual models. The update also adds deferred tool loading, cache warming for Anthropic models, mid-conversation system messages that adapt prompts and tools based on transcript context, a new terminal user interface (TUI) theme, and full-screen mode by default.
Pi Durable ports the Pi agent harness to TypeScript and externalizes all stateful components. This enables crash survival through checkpointed tasks, allowing agents and subagents to resume from their last exact state after a process failure or restart.
The TypeScript port supports pluggable storage backends, including Memory, SQLite, and JSONL, and runs in any JavaScript runtime such as Node, Bun, or Cloudflare Workers. It also supports flexible remote or local execution environments.
Pi Durable introduces concurrency features that allow a single harness to run multiple parallel, branching conversations—such as a main channel and separate threads—without blocking one another.
Developers can bundle custom system prompts, tools, hooks, and durable tasks into installable “Extensions.” Pi Durable also supports dynamic hot-swapping of tool and extension code while the agent is running, with the next tool call automatically picking up the new code.
Context management in Pi Durable automatically compacts older messages to maintain token limits without pausing the agent’s active work. Application state, such as a to-do list, is stored in documents alongside conversation transcripts, enabling multiplayer and state synchronization.
This allows multiple users or UIs to connect, watch, and steer the same agent simultaneously.
- Oct 14, 2026 · Latent Space — swyx
Google releases Gemini 4 Argon globally with agentic coding claims and cost schedule
Trust74 - Oct 14, 2026 · Latent Space — swyx
Dyna Robotics releases Dyna-2.1, a physical agent with hour-long autonomy using a world-action model trained on one million hours of video
Trust74 - Oct 13, 2026 · Latent Space — swyx
Anthropic introduces dynamic workflows for Claude Managed Agents to coordinate multi-agent orchestration
Trust72