[Submission] Pi Durable on Temporal #145

Activity

1. Shy

   added

   code exchange submissionCode and/or content about Temporal!Code and/or content about Temporal!

   triageIssues that Temporal folk need to look atIssues that Temporal folk need to look at

   on Oct 8, 2026
2. github-actions commented on Oct 8, 2026

   

   github-actionsbot

   on Oct 8, 2026 – with GitHub Actions

   More actions

   ZiggyBot — automated pre-evaluation

   ---

   Criterion	Assessment
   Useful to Temporal users	✅ Yes — Provides a Temporal backend for Pi Durable's agent harness, running the harness in a Workflow with model calls and tools as Activities.
   Benefits clearly explained	✅ Yes — README and linked docs describe the integration model, usage, and stated limitations.
   OSI Approved License	✅ Yes — MIT license is present and identified.
   README with working instructions	⚠️ Unclear — Install and development steps are present, but installation requires a GitHub Packages classic token and the usage section defers details to linked docs not included here.

   Overall: Needs review ⚠️

   Notes:

   • Scope is explicitly experimental; author reports testing of model calls, filesystem tools, worker restart, Continue-as-New, and history replay on Temporal Cloud, with shared S3 storage untested.
   • The native-client example imports from a local path (../../.local/pi-upstream/...) rather than a published package, which suggests that example depends on a local checkout of Pi's upstream experimental client and may not build from a clean clone without additional setup.
   • Install requires authenticating to GitHub Packages with a classic token; this adds friction and may block users who expect public npm distribution.
   • Determinism controls are present: use of patched() for versioned behavior changes, continueAsNew, and allHandlersFinished, consistent with replay-safe Workflow evolution.
   • Workflow-to-Workflow communication is used in the agent-harness example via getExternalWorkflowHandle and signaling a parent Workflow with trace events.

   Suggested questions for submitter:

   • Does the native-client example build from a clean clone, or does it require a local checkout of Pi's upstream repository at .local/pi-upstream?
   • Is public (unauthenticated) distribution planned, or is GitHub Packages with a token the intended install path?
   • Which specific Pi harness operations are mapped to Activities versus run inline in the Workflow?

   Notable Temporal pattern:

   Use of patched() to gate behavior changes (model selection, session rollover) so that in-flight Workflow histories replay deterministically across code versions. This is representative because the project runs a long-lived agent harness in a Workflow and relies on replay safety across worker restarts and Continue-as-New.
   View in source ↗

   ---

   Automated AI pre-screen by ZiggyBot. Final decisions are made by the community team.

   Reviewer: Check the box below to generate a Contentful-ready long description for this submission.

   • Generate long description if needed
3. github-actions

   added

   ziggy reviewedPre-screened by ZiggyBotPre-screened by ZiggyBot

   on Oct 8, 2026
4. h0rv commented on Oct 9, 2026

   

   h0rv

   on Oct 9, 2026

   Author

   More actions

   The native client example works from a clean clone after npm ci --ignore-scripts and npm run example:client:setup. Setup checks out the pinned Pi revision into .local/pi-upstream and installs its dependencies. CI runs that setup before checking the native client example. The adapter package doesn't require the checkout.

   Version 0.1.0 is now on public npm. Install with npm install @h0rv/pi-durable-temporal. I verified a fresh install without authentication.

   Pi's harness, conversation state and hooks run inside the Workflow. Model requests and external tool operations run as Activities. Tools that need Pi's live conversation or task APIs stay in the Workflow and delegate their external work to Activities. The MCP example runs Code Mode as one Activity, including its nested MCP calls.