Perplexity 详细介绍了其新的 Perplexity Projects 功能背后的技术架构，该功能允许 AI 代理协作处理长期任务。该系统利用 Git 作为基础元素，将其视为代理提交工作的熔炉。每个代理都在一个短暂的沙箱中运行，其进度存储在 S3 支持的存储库中以防止数据丢失。为了管理并发工作并维护项目完整性，Perplexity 采用了提交的短期锁定系统，并确保新的提交建立在最新的接受版本之上，代理在必要时重放更改。 AI
影响 解释了协作式 AI 代理的技术基础，可能会影响未来多代理系统的构建方式。
排序理由 对产品功能的详细技术解释，而非新产品发布或前沿发布。
AI 生成摘要 · Google Gemini · 来自 6 个来源。 我们如何撰写摘要 →
完整方法见我们的编辑标准。
Because each sandbox holds a real checkout, agents search it with grep and standard file tools instead of relying on broad semantic searches. On an internal benchmark of 150 financial-document questions, this reached 0.627 answer accuracy versus about 0.37 for semantic search. h…
A single long-lived sandbox per project would need provisioning for peak load and months of uptime. Instead, repositories live on S3-backed storage and each agent runs in its own ephemeral sandbox, so a crash loses only unsubmitted work. https://t.co/dglC3ZI6AX
The CLI holds a short-lived lock to serialize submits, and the Git service accepts a submit only if it builds on the head, the latest accepted commit. Otherwise, the agent replays its change on the new head and resubmits. https://t.co/urIrVVwPL7
Git provides atomic multi-file commits, authorship, and queryable history. A forge hosts the repository, controls who can update it, and adds PR review. Projects keep the first two and replace review with a check of each submit against a repository contract.
Agents collaborating on a long-running project face the same problem as engineers sharing a codebase. Work must outlive ephemeral sandboxes, concurrent writes must not overwrite each other, and history must show who changed what.
This is how we built the filesystem behind Perplexity Projects as a Git forge for agents. Each agent works in its own sandbox and commits to a shared repository where people and other agents can build on its work across sessions. Read more: https://t.co/jcrTptb2IW https://t.co/…