跳到正文
原文
Hacker News· hypervs·· 2 天前精选AI 评分48

Trigora 发布基于 TCC 引擎的 durable 执行平台

Show HN: Trigora – durable execution without history replay

AI 导读

Trigora 推出基于 Transparent Continuation Checkpointing (TCC) 的 durable 执行平台,面向长生命周期 AI 智能体和程序。

推荐理由

原文给出 TCC 与传统历史回放方案的差异,并附有快速上手的安装命令,便于评估长生命周期 AI 智能体执行框架时直接对照。

正文

Trigora — durable execution without history replay.

npm version npm downloads License: MIT GitHub stars

TypeScript · Python · Rust · TCC Engine · Cloud

Durable execution without history replay.

Trigora is a durable execution platform for long-lived AI agents and programs.

Write normal application logic in TypeScript, Python, or Rust. Trigora can suspend execution across events, timers, external effects, and child programs, then recover from committed continuation state after failure without replaying completed execution history.

import { effect, waitForEvent } from "@trigora/sdk";

export default async function approval() {
  const result = await effect("generate", () => 42);

  const review = await waitForEvent("approved");

  return { result, review };
}

Deploy it once. Start executions from the CLI, an API call, a webhook, or a cron trigger. Let them run for seconds, hours, or days.

Why Trigora?

Traditional durable workflow systems often reconstruct execution state by replaying prior history.

Trigora uses Transparent Continuation Checkpointing (TCC) instead.

At durable boundaries, TCC commits the program position and the live durable state required to continue. Recovery resumes from that committed continuation rather than re-executing the completed prefix.

That makes Trigora a natural fit for workloads that:

  • run long-lived AI agents;
  • call models, tools, APIs, and external systems;
  • wait for humans or external events;
  • sleep for minutes, hours, or days;
  • coordinate durable child executions;
  • branch dynamically;
  • need to survive worker or process failure.

How TCC works →
Read the technical report →

Start building

Install the SDK, client, and CLI for your language.

TypeScript

npm install @trigora/sdk @trigora/client trigora

Python

pip install trigora trigora-client trigora-cli

Requires Python 3.10–3.14.

Rust

cargo add trigora trigora-client
cargo install trigora-cli

Then initialize a project:

trigora init

Start the local runtime:

trigora dev

When you're ready to run in production:

trigora deploy

Read the quickstart →

Trigora Cloud

Trigora Cloud is the managed production runtime for TCC programs.

It provides:

  • durable execution;
  • deployment and program versioning;
  • events and timers;
  • cron and webhook triggers;
  • project secrets;
  • execution inspection and observability;
  • workspace access and API tokens;
  • managed persistence and recovery.

Start at cloud.trigora.dev.

Repository map

Trigora is split into focused repositories.

Repository What it contains
trigora CLI, local runtime, public contracts, and ecosystem entry point
trigora-typescript TypeScript authoring SDK and API client
trigora-python Python authoring SDK and API client
trigora-rust Rust authoring SDK and API client
tcc-engine Portable TCC execution engine, language frontends, host protocol, and specifications

This repository

trigora-dev/trigora contains the pieces shared across the Trigora ecosystem:

  • the trigora CLI;
  • the native local runtime used by trigora dev;
  • shared public contracts;
  • release coordination for the Trigora SDK ecosystem.

Language-specific authoring APIs live in their respective repositories above.

TCC Engine

TCC Engine is the portable execution technology underneath Trigora.

It is implemented primarily in Rust, supports native and WebAssembly embedding, and has declared frontends for TypeScript, Python, and Rust.

The engine is available separately for teams that want to embed TCC into their own runtimes or infrastructure.

TCC Engine on GitHub →

TCC Engine is source-available under the Business Source License 1.1 and converts to Apache 2.0 under its license terms. The Trigora SDKs, clients, CLI, and public contracts are MIT licensed.

Research

The public research covers:

  • continuation-based recovery;
  • matched recovery versus history replay;
  • live-state scaling;
  • crash recovery and fresh-process restore;
  • frontend semantics;
  • host responsibilities;
  • limitations and failure windows.

Technical report →

Links

Company

Trigora is built by Trigora, Inc.

License

MIT © 2026 Trigora, Inc.

来源:Hacker News · github.com