Typeflux · Typed AI activities + workflows

Typed AI workflows in Python and TypeScript.

Turn model calls into workflows you can test, adapt and understand. Keep your types, prompts and providers working together, with evidence behind every result.

Connect your coding agent Interactive Field Guide

Make the contract executable.

Describe what goes in, what comes out and how the work connects. Typeflux brings the prompts, model calls and validation together.

from typeflux.core import PromptRef, ai_activity

import { defineActivity } from "@typeflux/temporal";

These are activity authoring APIs. The maintained Local quickstarts provide complete YAML, schemas and scripted fake providers for credential-free CLI and async SDK execution. Real model calls require provider configuration, optional dependencies and credentials.

AI workflows drift across more than code.

Schemas, prompts, providers, workflow code, environments, policy and execution can move at different speeds. Their evidence needs to stay connected.

Keep your AI workflows in sync.

Models change. Prompts evolve. Your application still needs a clear contract. Typeflux keeps those pieces connected: schemas define the contract, prompts stay adaptable, and evidence travels with the work.

One contract across the workflow lifecycle.

Build typed AI activities and composable workflows. Test with scripted providers, keep components portable, and bring moderation, redaction and operational evidence into the application.

Keep the workflow. Change the stack.

Reuse supported computation across Local and Temporal. Swap Inline, LangSmith, Langfuse or custom prompt registries; OpenAI, Anthropic, Gemini or custom model providers; and observability backends. Typeflux surfaces capability differences. SDK-specific module wiring does not port automatically.

From prompt to validated result.

Validate input → resolve and render prompt → structured provider call → check and repair output → hook and moderation → cache, trace and return. The same activity pipeline powers both runtimes, with hooks, moderation, caching, redaction and observability.

Two runtimes. The same AI activity pipeline.

Run in your application or across durable workers. The same typed activities, prompts and providers power both—in Python or TypeScript.

Local

Foreground execution through the CLI or an awaited async SDK call. Use it for scripts, batch jobs and production applications with YAML sequences, conditions, parallel branches, bounded ordered maps and same-project children. No Temporal SDK, service, worker daemon or workflow database is required.

Run Local · Python · Run Local · TypeScript

Temporal

Durable history, configured recovery and activity retries/timeouts, timers, signals, queries, human review, configured compensation and managed run operations. Requires workers and a reachable Temporal service.

Run Temporal · Python · Run Temporal · TypeScript

How execution, recovery and setup work

Local is the default YAML runtime on current main. Temporal requires explicit runtime.engine: temporal and a task queue.

Local loses progress on process exit. A fresh invocation starts from the beginning and can repeat completed effects. Local traces and correlation IDs are evidence, not resumable execution history.

Local run ownership remains CLI/SDK-only. Console, control-plane and MCP run operations are Temporal-backed; authoring tools and exported observations do not manage Local executions.

Moving supported computation to Temporal starts a new execution; it does not migrate an in-flight Local run. Neither runtime guarantees exactly-once external side effects. Design for idempotency and reconciliation.

Runtime guide · Management boundary

Local examples target current source and candidate packages. Install the actual candidate wheel or tarballs using the quickstarts. Published Python 0.1.0 predates Local; @typeflux/yaml is not yet on npm. Registry availability checked November 12, 2026.

Your language. Your workflow.

Build with Python or TypeScript, give your coding agent shared context, and see what’s coming for team operations. One framework connects your code, tools and workflows. Both SDKs are Alpha.

Python documentation · TypeScript documentation

Compare SDK editions and current gaps

Typeflux MCP

Give your coding agent the same contracts, examples and project context you use. Through MCP, your agent can inspect schemas and scaffold a change. From the terminal, it can run a Local workflow with the Python or TypeScript CLI and check its JSON result. Connect your editor with Typeflux MCP using the maintained setup guide.

In active development

Typeflux Cloud Console

Bring your team’s Temporal deployments into one view, from workflow topology and run state to review decisions and policy evidence.

Sign up to learn more

Bring a real workflow.

Tell us what you are building and what would make it easier to ship. Connect with the Typeflux team or request early access to the Cloud Console, currently in active development. The form requires JavaScript.