Show HN: IntentusNet – 基于 WAL 的确定性回放，用于 AI 工具执行
1 分•作者: balachandarmani•6 个月前
大家好——我一直在开发 IntentusNet，这是一个小型执行运行时，专注于围绕 AI 工具的确定性、可重放的执行语义。<p>我在生产系统中一直遇到的问题是：
AI 管道是可观察的，但不可重现。发生事件后，模型、路由逻辑、重试或回退可能已经改变——仅凭日志无法重现实际发生的情况。<p>v1.3.0 引入了运行时确定性核心：<p>先写日志（仅追加 JSONL），在副作用之前写入<p>崩溃安全恢复和确定性重放（在出现差异时大声失败）<p>运行时执行合约（超时、重试、成本上限）<p>副作用分类，以防止不安全的重试或回退<p>CLI 优先检查（列表 / 显示 / 跟踪 / 重放 / 差异）<p>它不是一个规划器或代理框架，也不是 MCP 的替代品——它纯粹专注于围绕工具（包括 MCP 风格的工具）的执行语义。<p>快速尝试（从 repo 根目录运行）：<p>git clone <a href="https://github.com/Balchandar/intentusnet" rel="nofollow">https://github.com/Balchandar/intentusnet</a><p>cd intentusnet
pip install -e .
python -m examples.deterministic_routing_demo.demo --mode with
python -m examples.deterministic_routing_demo.demo --mode mcp<p>文档（架构、保证、演示）：
<a href="https://intentusnet.com" rel="nofollow">https://intentusnet.com</a><p>MIT 许可，开源：
<a href="https://github.com/Balchandar/intentusnet" rel="nofollow">https://github.com/Balchandar/intentusnet</a><p>我非常重视来自构建真实系统的人的反馈：<p>您在实践中对确定性重放有什么期望？<p>您如何在 AI 管道中安全地处理重试和副作用？
查看原文
Hi folks — I’ve been working on IntentusNet, a small execution runtime that focuses on deterministic, replayable execution semantics around AI tools.<p>The problem I kept hitting in production systems:
AI pipelines are observable, but not reproducible. After an incident, models, routing logic, retries, or fallbacks may have changed — logs alone don’t let you replay what actually happened.<p>v1.3.0 introduces a runtime determinism core:<p>Write-ahead log (append-only JSONL) written before side effects<p>Crash-safe recovery and deterministic replay (fails loud on divergence)<p>Runtime execution contracts (timeouts, retries, cost ceilings)<p>Side-effect classification to prevent unsafe retries or fallback<p>CLI-first inspection (list / show / trace / replay / diff)<p>It’s not a planner or agent framework, and not a replacement for MCP — it focuses purely on execution semantics around tools (including MCP-style tools).<p>Quick try (run from repo root):<p>git clone <a href="https://github.com/Balchandar/intentusnet" rel="nofollow">https://github.com/Balchandar/intentusnet</a><p>cd intentusnet
pip install -e .
python -m examples.deterministic_routing_demo.demo --mode with
python -m examples.deterministic_routing_demo.demo --mode mcp<p>Docs (architecture, guarantees, demos):
<a href="https://intentusnet.com" rel="nofollow">https://intentusnet.com</a><p>MIT licensed, open source:
<a href="https://github.com/Balchandar/intentusnet" rel="nofollow">https://github.com/Balchandar/intentusnet</a><p>I’d really value feedback from people building real systems:<p>What guarantees do you expect from deterministic replay in practice?<p>How do you handle retries and side effects safely in AI pipelines?