Pi’s coding-agent harness has reached version 1.0, adding native Model Context Protocol support and a fuller set of tools for developers. Its creators have also introduced Pi Durable, an experimental framework for building agents that can resume work after a process fails and run multiple conversations in parallel.
Watch Desk analysis
What happened
The Register describes Pi 1.0 as the first stable release of the open-source, MIT-licensed terminal coding agent. The release adds MCP support, alongside changes to its interface, extensions and tool use. Read The Register’s report.
Pi Durable is a separate framework, not simply another button in the terminal agent. Latent Space’s AINews account describes it as a TypeScript port that externalises Pi’s stateful components, with checkpointed tasks, pluggable storage and support for parallel conversations.
Our top picks
- Native MCP support
Pi can connect with tools and services built around the Model Context Protocol. - Codemode and virtual model extensions
The release adds Codemode functionality and lets developers extend support for virtual models. - Deferred tool loading
Tools can be loaded when needed rather than all being brought into the conversation up front. - Mid-conversation system messages
Developers can change system prompts and tools while keeping the conversation transcript in view. - Pi Durable checkpoints
Its experimental framework records task steps so work can resume from the last saved state after a process failure. - Parallel conversations
Pi Durable can run branching conversations without making one wait for another.
Why it matters
MCP support gives Pi a route into a growing ecosystem of external tools, while the other changes offer more control over how a coding agent uses context and extensions. Pi Durable tackles a different, practical headache: keeping longer-running agent tasks alive and manageable when a process stops or several workstreams need to run at once.
That is a substantial step beyond tidying up a terminal interface, but Pi Durable is experimental. The described recovery and concurrency features are design aims and reported capabilities, not a guarantee that every interrupted task will pick up perfectly.
Our read
Pi 1.0 makes the coding harness more capable without losing its open-source, terminal-first character. Pi Durable is the more ambitious idea: it treats an agent’s progress as state that can be saved, moved and shared, rather than a session that vanishes when the process does. Worth exploring, but the stable release and experimental framework should not be mistaken for the same level of maturity.
What to watch
- Which MCP integrations Pi users put to work in practice.
- How Pi Durable handles failures and state recovery across different storage backends.
- Whether the experimental framework develops into a dependable foundation for long-running agent tasks.
Discussion spark: Which matters more for coding agents: a richer tool ecosystem, or durable state that survives when the process stops?
Sources and evidence
- Pi coding agent pulls a 180 and adds MCP support (2 October 2026, 07:01 UTC)
- [AINews] Pi 1.0, Pi Durable, and AIE NYC (2 October 2026, 07:01 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.