放出去 · 派一批活上云,验收回来Send It Out · Dispatch a Batch to the Cloud, Review What Comes Back.
有些活不用你守着。这章把一批活真的派出去:两个并行的 cloud agent 在后台跑,你去喝咖啡,回来照着 transcript 逐条验收;跑顺的那个改成每周定时。也讲清对一个人,多 agent 编排什么时候是过度。Some work doesn't need you watching. This chapter actually sends a batch out: two parallel cloud agents run in the background while you get coffee, then you review them line by line against the transcripts; the one that worked becomes a weekly schedule. Plus: when multi-agent orchestration is overkill for one person.
到这里 agent 都在你眼前跑 —— 你看着它一步步动。但有些活不值得你守着:夜里的维护、一开 PR 就该触发的检查、需要并行跑的一批杂活。这章把一批活真的派出去:两个 cloud agent 后台并行,你去喝咖啡,回来照着 transcript 和 diff 逐条验收;跑顺的那个,改成每周定时。然后讲清:对一个人,多 agent 编排什么时候是过度。So far the agent has run in front of you — you watch it move step by step. But some work isn't worth watching: overnight maintenance, a check that should fire the moment a PR opens, a batch of chores that could run in parallel. This chapter actually sends a batch out: two cloud agents running in the background while you get coffee, then you review them line by line against transcripts and diffs; the one that worked becomes a weekly schedule. Then: when multi-agent orchestration is overkill for one person.
— I
主线:派两个 cloud agent 出去,验收回来The Main Line: Send Two Cloud Agents Out, Review What Comes Back.
挑两件安全又能并行验收的活:一件出报告(依赖审计),一件出 PR(清一组 lint 警告)。两个都是明确、可验证、不需要你中途判断的活 —— 这是上云的入场条件。跑 cloud agents 需要账户里 ≥20 credits(新用户有试用 credits);Slack / Linear 这类 integrations 要 Build / Max / Business 的团队,本章只用定时和手动,不要团队。2注 2Note 2Warp Docs · Cloud agents quickstart —— 截至 2026-08-06:入口是 /cloud-agent slash 命令(没有 environment 会引导你建;/create-environment 显式建);environment = Docker image + repos + setup commands(要求 glibc,不支持 Alpine / musl;有预置 dev images;GitHub auth 可按用户或按团队配);跑 cloud agents 需要 ≥20 credits(新用户有试用 credits);run 在 conversations 面板、session 链接、或 Oz web app(oz.warp.dev)的 Runs tab 看,手机也行;/create-skill 把一次 run 变成可复用、可调度的 skill;oz schedule create 建定时;Slack / Linear 等 integrations 需要 Build / Max / Business 的团队。Warp Docs · Cloud agents quickstart — as of 2026-08-06: the entry point is the /cloud-agent slash command (it guides environment creation if missing; /create-environment builds one explicitly); an environment = Docker image + repos + setup commands (glibc required — no Alpine / musl; prebuilt dev images available; GitHub auth per user or team-level); running cloud agents requires ≥20 credits (new users get trial credits); view runs in the conversations panel, via the session link, or in the Oz web app (oz.warp.dev) Runs tab — mobile too; /create-skill turns a run into a reusable, schedulable skill; oz schedule create makes schedules; integrations like Slack / Linear require a team on Build / Max / Business.Pick two safe jobs with checkable outputs: one produces a report (a dependency audit), one produces a PR (clearing a set of lint warnings). Both are well-defined, verifiable, and need no mid-run judgment — that's the admission ticket for cloud. Cloud agents need ≥20 credits in the account (new users get trial credits); integrations like Slack / Linear require a team on Build / Max / Business — this chapter uses only schedules and manual runs, no team needed.2注 2Note 2Warp Docs · Cloud agents quickstart —— 截至 2026-08-06:入口是 /cloud-agent slash 命令(没有 environment 会引导你建;/create-environment 显式建);environment = Docker image + repos + setup commands(要求 glibc,不支持 Alpine / musl;有预置 dev images;GitHub auth 可按用户或按团队配);跑 cloud agents 需要 ≥20 credits(新用户有试用 credits);run 在 conversations 面板、session 链接、或 Oz web app(oz.warp.dev)的 Runs tab 看,手机也行;/create-skill 把一次 run 变成可复用、可调度的 skill;oz schedule create 建定时;Slack / Linear 等 integrations 需要 Build / Max / Business 的团队。Warp Docs · Cloud agents quickstart — as of 2026-08-06: the entry point is the /cloud-agent slash command (it guides environment creation if missing; /create-environment builds one explicitly); an environment = Docker image + repos + setup commands (glibc required — no Alpine / musl; prebuilt dev images available; GitHub auth per user or team-level); running cloud agents requires ≥20 credits (new users get trial credits); view runs in the conversations panel, via the session link, or in the Oz web app (oz.warp.dev) Runs tab — mobile too; /create-skill turns a run into a reusable, schedulable skill; oz schedule create makes schedules; integrations like Slack / Linear require a team on Build / Max / Business.
01
备环境:让 /cloud-agent 带你建一个Prepare the environment: let /cloud-agent walk you through one
在 Warp 里输入 /cloud-agent,没有 environment 它会引导你建:一个 environment 就是一个 Docker image + 你的 repo + setup 命令(装依赖、配 GitHub auth)。两条硬要求:镜像必须是 glibc 的(Alpine / musl 不行),不想写 Dockerfile 就用预置的 dev image。2注 2Note 2Warp Docs · Cloud agents quickstart —— 截至 2026-08-06:入口是 /cloud-agent slash 命令(没有 environment 会引导你建;/create-environment 显式建);environment = Docker image + repos + setup commands(要求 glibc,不支持 Alpine / musl;有预置 dev images;GitHub auth 可按用户或按团队配);跑 cloud agents 需要 ≥20 credits(新用户有试用 credits);run 在 conversations 面板、session 链接、或 Oz web app(oz.warp.dev)的 Runs tab 看,手机也行;/create-skill 把一次 run 变成可复用、可调度的 skill;oz schedule create 建定时;Slack / Linear 等 integrations 需要 Build / Max / Business 的团队。Warp Docs · Cloud agents quickstart — as of 2026-08-06: the entry point is the /cloud-agent slash command (it guides environment creation if missing; /create-environment builds one explicitly); an environment = Docker image + repos + setup commands (glibc required — no Alpine / musl; prebuilt dev images available; GitHub auth per user or team-level); running cloud agents requires ≥20 credits (new users get trial credits); view runs in the conversations panel, via the session link, or in the Oz web app (oz.warp.dev) Runs tab — mobile too; /create-skill turns a run into a reusable, schedulable skill; oz schedule create makes schedules; integrations like Slack / Linear require a team on Build / Max / Business.Type /cloud-agent in Warp; if you have no environment it guides you through one: an environment is a Docker image + your repo + setup commands (install deps, configure GitHub auth). Two hard requirements: the image must be glibc-based (no Alpine / musl), and if you don't want to write a Dockerfile, use a prebuilt dev image.2注 2Note 2Warp Docs · Cloud agents quickstart —— 截至 2026-08-06:入口是 /cloud-agent slash 命令(没有 environment 会引导你建;/create-environment 显式建);environment = Docker image + repos + setup commands(要求 glibc,不支持 Alpine / musl;有预置 dev images;GitHub auth 可按用户或按团队配);跑 cloud agents 需要 ≥20 credits(新用户有试用 credits);run 在 conversations 面板、session 链接、或 Oz web app(oz.warp.dev)的 Runs tab 看,手机也行;/create-skill 把一次 run 变成可复用、可调度的 skill;oz schedule create 建定时;Slack / Linear 等 integrations 需要 Build / Max / Business 的团队。Warp Docs · Cloud agents quickstart — as of 2026-08-06: the entry point is the /cloud-agent slash command (it guides environment creation if missing; /create-environment builds one explicitly); an environment = Docker image + repos + setup commands (glibc required — no Alpine / musl; prebuilt dev images available; GitHub auth per user or team-level); running cloud agents requires ≥20 credits (new users get trial credits); view runs in the conversations panel, via the session link, or in the Oz web app (oz.warp.dev) Runs tab — mobile too; /create-skill turns a run into a reusable, schedulable skill; oz schedule create makes schedules; integrations like Slack / Linear require a team on Build / Max / Business.
02
派:两条任务并行发出去Dispatch: send both tasks out in parallel
用下面 PromptBox 的模板填两条任务,各起一个 cloud run。命令行派用 oz agent run-cloud,app 里派就用 /cloud-agent 开 Cloud Mode。发完关上笔记本 —— 它们不占你的终端。Fill in two tasks from the PromptBox template below and start one cloud run each. From the command line use oz agent run-cloud; in the app, open Cloud Mode with /cloud-agent. Then close the lid — they don't occupy your terminal.
03
离开:run 的状态随处可看Walk away: run status is visible anywhere
每个 run 留下持久记录:状态、元数据、transcript、outputs。在 conversations 面板、session 链接、或 oz.warp.dev 的 Runs tab 都能看 —— 手机上也能盯。这一步是真的去喝咖啡,不是换个窗口焦虑。1注 1Note 1Warp Docs · Oz Platform overview —— 截至 2026-08-06:Cloud Agents 运行在 Oz Platform 上 —— trigger(schedule、Slack / Linear / GitHub、CI、webhook、API、手动)创建任务,编排层(Warp cloud 控制面)跟踪生命周期,任务在 host(默认 Warp 托管;self-hosted 是 Enterprise)上执行,可选 environment;每个 run 留下 status、metadata、transcript、outputs 的持久记录。Oz CLI 是 headless 接口;另有 Agent API 与 Python / TypeScript SDK。Warp Docs · Oz Platform overview — as of 2026-08-06: Cloud Agents run on the Oz Platform — triggers (schedule, Slack / Linear / GitHub, CI, webhook, API, manual) create tasks, the orchestration layer (Warp's cloud control plane) tracks the lifecycle, tasks execute on a host (Warp-hosted by default; self-hosted is Enterprise), optionally in an environment; every run leaves a persistent record — status, metadata, transcript, outputs. The Oz CLI is the headless interface; there's also an Agent API and Python / TypeScript SDKs.Every run leaves a persistent record: status, metadata, transcript, outputs. Watch from the conversations panel, the session link, or the Runs tab at oz.warp.dev — your phone works too. This step is actually getting coffee, not anxiety in another window.1注 1Note 1Warp Docs · Oz Platform overview —— 截至 2026-08-06:Cloud Agents 运行在 Oz Platform 上 —— trigger(schedule、Slack / Linear / GitHub、CI、webhook、API、手动)创建任务,编排层(Warp cloud 控制面)跟踪生命周期,任务在 host(默认 Warp 托管;self-hosted 是 Enterprise)上执行,可选 environment;每个 run 留下 status、metadata、transcript、outputs 的持久记录。Oz CLI 是 headless 接口;另有 Agent API 与 Python / TypeScript SDK。Warp Docs · Oz Platform overview — as of 2026-08-06: Cloud Agents run on the Oz Platform — triggers (schedule, Slack / Linear / GitHub, CI, webhook, API, manual) create tasks, the orchestration layer (Warp's cloud control plane) tracks the lifecycle, tasks execute on a host (Warp-hosted by default; self-hosted is Enterprise), optionally in an environment; every run leaves a persistent record — status, metadata, transcript, outputs. The Oz CLI is the headless interface; there's also an Agent API and Python / TypeScript SDKs.
04
验收:照 transcript 逐条过Review: go line by line against the transcript
回来别只看结论。报告型任务:开 artifact,抽两条漏洞去 npm advisory 核对。PR 型任务:看 diff 全貌,跑一遍 CI,再问一句「它为什么这样改」—— transcript 里有它的推理。不合格就拒,这就是验收。Don't just read conclusions when you're back. For the report task: open the artifact, spot-check two vulnerabilities against npm advisories. For the PR task: read the full diff, let CI run, and ask why it changed what it changed — the transcript holds its reasoning. Reject what fails. This is the review.
05
固化:跑顺的那个改成每周定时Fix in place: the one that worked becomes a weekly schedule
依赖审计跑顺了,就别再手动派:oz schedule create --cron "0 10 * * 1" 每周一上午自动跑;或者 /create-skill 把这次 run 变成可复用、可调度的 skill。lint 那个如果只发生一次,就让它停留在一次性 —— 不是什么都值得定时。2注 2Note 2Warp Docs · Cloud agents quickstart —— 截至 2026-08-06:入口是 /cloud-agent slash 命令(没有 environment 会引导你建;/create-environment 显式建);environment = Docker image + repos + setup commands(要求 glibc,不支持 Alpine / musl;有预置 dev images;GitHub auth 可按用户或按团队配);跑 cloud agents 需要 ≥20 credits(新用户有试用 credits);run 在 conversations 面板、session 链接、或 Oz web app(oz.warp.dev)的 Runs tab 看,手机也行;/create-skill 把一次 run 变成可复用、可调度的 skill;oz schedule create 建定时;Slack / Linear 等 integrations 需要 Build / Max / Business 的团队。Warp Docs · Cloud agents quickstart — as of 2026-08-06: the entry point is the /cloud-agent slash command (it guides environment creation if missing; /create-environment builds one explicitly); an environment = Docker image + repos + setup commands (glibc required — no Alpine / musl; prebuilt dev images available; GitHub auth per user or team-level); running cloud agents requires ≥20 credits (new users get trial credits); view runs in the conversations panel, via the session link, or in the Oz web app (oz.warp.dev) Runs tab — mobile too; /create-skill turns a run into a reusable, schedulable skill; oz schedule create makes schedules; integrations like Slack / Linear require a team on Build / Max / Business.If the dependency audit ran clean, stop dispatching it by hand: oz schedule create --cron "0 10 * * 1" runs it every Monday morning; or /create-skill turns the run into a reusable, schedulable skill. The lint one, if it only happens once, stays one-off — not everything deserves a schedule.2注 2Note 2Warp Docs · Cloud agents quickstart —— 截至 2026-08-06:入口是 /cloud-agent slash 命令(没有 environment 会引导你建;/create-environment 显式建);environment = Docker image + repos + setup commands(要求 glibc,不支持 Alpine / musl;有预置 dev images;GitHub auth 可按用户或按团队配);跑 cloud agents 需要 ≥20 credits(新用户有试用 credits);run 在 conversations 面板、session 链接、或 Oz web app(oz.warp.dev)的 Runs tab 看,手机也行;/create-skill 把一次 run 变成可复用、可调度的 skill;oz schedule create 建定时;Slack / Linear 等 integrations 需要 Build / Max / Business 的团队。Warp Docs · Cloud agents quickstart — as of 2026-08-06: the entry point is the /cloud-agent slash command (it guides environment creation if missing; /create-environment builds one explicitly); an environment = Docker image + repos + setup commands (glibc required — no Alpine / musl; prebuilt dev images available; GitHub auth per user or team-level); running cloud agents requires ≥20 credits (new users get trial credits); view runs in the conversations panel, via the session link, or in the Oz web app (oz.warp.dev) Runs tab — mobile too; /create-skill turns a run into a reusable, schedulable skill; oz schedule create makes schedules; integrations like Slack / Linear require a team on Build / Max / Business.
提示词PromptCloud Agent 任务模板(主线两件活都照这个填)Cloud Agent task template (fill both main-line jobs with this)
任务:<后台要完成什么>
环境:<repo / image / setup commands>
边界:只允许 <读/写范围>;禁止 <生产写入/删除/deploy/secret 输出>
验收:<artifact / PR / 报告>,附上你跑过的验证命令和结果
失败:失败时停止,保留 logs,不自动重试破坏性命令。Task: <what should finish in the background>
Environment: <repo / image / setup commands>
Boundary: allow only <read/write scope>; forbid <production write/delete/deploy/secret output>
Acceptance: <artifact / PR / report>, with the verification commands you ran and their results
Failure: stop on failure, preserve logs, do not auto-retry destructive commands.
任务:跑一次依赖审计,找出 npm 里有已知漏洞或落后两个大版本的包
环境:repo github.com/acme/web,node:22 image,setup 跑 npm ci
边界:只允许读 package.json / package-lock.json 和跑 npm audit;禁止生产写入、删除、deploy、secret 输出
验收:一份按严重度排序的报告 artifact,附升级建议命令,说明每条结论来自哪条 advisory
失败:失败时停止,保留 logs,不自动重试破坏性命令。Task: run a dependency audit to find npm packages with known vulnerabilities or two major versions behind
Environment: repo github.com/acme/web, node:22 image, setup runs npm ci
Boundary: allow only reading package.json / package-lock.json and running npm audit; forbid production writes, deletes, deploys, secret output
Acceptance: a report artifact sorted by severity, with suggested upgrade commands, citing the advisory behind each conclusion
Failure: stop on failure, preserve logs, do not auto-retry destructive commands.
命令行派活与定时(oz CLI;app 里对应 /cloud-agent 与 schedule UI)Dispatching and scheduling from the CLI (oz; in the app, /cloud-agent and the schedule UI)
— II
本地起,云端接;云端完,本地审Start Local, Hand to Cloud; Finish Cloud, Review Local.
你不必一开始就决定本地还是云端 —— 一段对话可以跑着跑着交接。三个方向规则不同:本地 → 云端只有 Warp Agent 能走,fork 会话并把你未提交的改动快照带上去;云端 → 云端(Warp Agent、Claude Code、Codex harness 之间)同一会话继续,workspace 状态恢复;云端 → 本地用 Continue locally / /continue-locally,但要记住:workspace 补丁不会应用到本地 —— 回来的东西去 branch / PR 里审,别指望工作区自动变。附件全程带得过去,补丁应用是 best-effort,部分失败会明说。3注 3Note 3Warp Docs · Handoff between local and cloud agents —— 截至 2026-08-06:本地 → 云端(仅 Warp Agent):fork 会话并应用未提交的 workspace 快照;云端 → 云端(Warp Agent 与 Claude Code / Codex harness):同一会话继续,workspace 状态恢复;云端 → 本地:Continue locally / /continue-locally,但 workspace 补丁不会应用到本地 —— 去审 branch / PR 这个 artifact;附件会带过去;补丁应用是 best-effort,部分失败会报告。Warp Docs · Handoff between local and cloud agents — as of 2026-08-06: local → cloud (Warp Agent only): forks the conversation and applies the uncommitted workspace snapshot; cloud → cloud (Warp Agent plus Claude Code / Codex harnesses): same conversation continues, workspace state restored; cloud → local: Continue locally / /continue-locally — but workspace patches do NOT apply locally, so review the branch / PR artifact instead; attachments carry over; patch application is best-effort with partial-failure reporting.You don't have to decide local or cloud up front — a conversation can hand off mid-run. The three directions follow different rules: local → cloud works for the Warp Agent only, forking the conversation and carrying your uncommitted workspace snapshot up; cloud → cloud (between Warp Agent, Claude Code, and Codex harnesses) continues the same conversation with workspace state restored; cloud → local uses Continue locally / /continue-locally, but remember: workspace patches do not apply locally — review what comes back as a branch / PR, don't expect your working tree to change on its own. Attachments carry over the whole way; patch application is best-effort and partial failures are reported.3注 3Note 3Warp Docs · Handoff between local and cloud agents —— 截至 2026-08-06:本地 → 云端(仅 Warp Agent):fork 会话并应用未提交的 workspace 快照;云端 → 云端(Warp Agent 与 Claude Code / Codex harness):同一会话继续,workspace 状态恢复;云端 → 本地:Continue locally / /continue-locally,但 workspace 补丁不会应用到本地 —— 去审 branch / PR 这个 artifact;附件会带过去;补丁应用是 best-effort,部分失败会报告。Warp Docs · Handoff between local and cloud agents — as of 2026-08-06: local → cloud (Warp Agent only): forks the conversation and applies the uncommitted workspace snapshot; cloud → cloud (Warp Agent plus Claude Code / Codex harnesses): same conversation continues, workspace state restored; cloud → local: Continue locally / /continue-locally — but workspace patches do NOT apply locally, so review the branch / PR artifact instead; attachments carry over; patch application is best-effort with partial-failure reporting.
选择Choice
适合Best for
不要用在Avoid for
本地 AgentLocal Agent
你要实时审命令、看输出、随时改方向You need to review commands, watch output, steer in real time
长时间等待、并行跑很多任务Long waits, many parallel tasks
Cloud AgentCloud Agent
明确、可验收、能后台跑、触发清楚Clear, reviewable, background task with a defined trigger
需要频繁人工判断、敏感本地凭据Frequent human judgment, sensitive local credentials
Oz orchestration
多 agent 并行、团队监督、run 可观察性、事件/计划自动化Multi-agent parallelism, team oversight, run observability, event/schedule automation
一个人临时跑一条命令A solo one-off command
— III
单人何时过度,账单怎么算When It's Overkill, and How the Bill Works.
多 agent 编排是给团队规模设计的 —— 一个人,一两个定时或手动的云端 agent 就够,再往上的 swarm 大多是过度,还多烧 credits。账单要会算:credits 分三桶 —— AI(推理)、compute(Warp 托管沙箱)、platform(run 生命周期与观测),同一池扣;每个 cloud run 至少扣 platform,用 Warp 托管算力加 compute,用 Warp 模型再加 AI。你本机配的 BYOK 云端拿不到 —— cloud run 照常消耗 Warp credits;Enterprise 要在云端用自己的推理,走 BYOLLM,不是本地那套 key。Cloud Runners(算力配置)2026-07-23 起全员可用,配 oz runner;但 self-hosted execution 仍是 Enterprise。Oz 还能把 Claude Code、Codex 当云端 harness 跑(pricing 页标 beta):推理记你的 provider 账,Warp 仍计 compute 与 platform —— 你已付费的 CLI 订阅上云接着用,但不是免费。4注 4Note 4Warp Docs · Credits —— 截至 2026-08-06:credits 分三桶:AI(推理)、compute(Warp 托管沙箱,仅 cloud run)、platform(run 生命周期 / integrations / 观测 —— 每个 cloud run 都扣,Business / Enterprise 用自带推理的本地 run 也扣);同一额度池(Settings > Billing and usage);每个 turn 有 credit chip 可 hover 看消耗;个人可以从自己的 credits 经 CLI / API 跑 cloud agents;Slack / Linear integrations 需要团队成员身份;普通 shell 命令不耗 credits。Warp Docs · Credits — as of 2026-08-06: three buckets — AI (inference), compute (Warp-hosted sandbox, cloud runs only), platform (run lifecycle / integrations / observability — charged on every cloud run, and on Business / Enterprise local runs using customer-supplied inference); one shared pool (Settings > Billing and usage); a per-turn credit chip shows usage on hover; individuals can run cloud agents from their own credits via CLI / API; Slack / Linear integrations require team membership; regular shell commands consume no credits.6注 6Note 6Warp Docs · BYOK + changelog 2026-07-23 —— BYOK 不适用于 Cloud Agents:key 只存在你设备上,云端 run 拿不到,cloud runs 照常消耗 Warp credits;旧的「Enterprise team-managed keys」例外已取消,集中托管推理走 BYOLLM。2026-07-23 起 Cloud Runners 与 Cloud Agent Runners 全员开放,配 oz runner CLI;2026-08-06 本机轻测:Warp bundle 带 oz(v0.2026.07.01),oz --help 列出 agent / environment / mcp / schedule / secret 等命令。Warp Docs · BYOK + the 2026-07-23 changelog — BYOK does not apply to Cloud Agents: keys live only on your device, cloud runs can't reach them and consume Warp credits as usual; the old 'Enterprise team-managed keys' exception is gone — centrally managed inference goes through BYOLLM. Since 2026-07-23, Cloud Runners and Cloud Agent Runners are open to all users, with the oz runner CLI; local light check on 2026-08-06: the Warp bundle ships oz (v0.2026.07.01), and oz --help lists agent / environment / mcp / schedule / secret and more.5注 5Note 5Warp Docs · Harnesses in Oz —— Claude Code 与 Codex 可作云端 harness,共享 triggers / environments / secrets / skills / rules / 观测;推理由你提供的凭据记 provider 的账,Warp 计 compute 与 platform credits;在 Cloud Mode 的 harness 下拉、Oz web app 或 API 的 harness 字段选;pricing 页把「Use any harness in the cloud」标为 beta(并列在 Free 下)。Warp Docs · Harnesses in Oz — Claude Code and Codex work as cloud harnesses, sharing triggers / environments / secrets / skills / rules / observability; inference is billed by your provider via credentials you supply, while Warp meters compute and platform credits; pick the harness in Cloud Mode's dropdown, the Oz web app, or the API's harness field; the pricing page labels 'Use any harness in the cloud' as beta (and lists it under Free).Multi-agent orchestration is built for team scale — solo, one or two scheduled or manual cloud agents are enough, and the swarm beyond that is mostly overkill that burns extra credits. Learn the bill: credits come in three buckets — AI (inference), compute (Warp-hosted sandbox), platform (run lifecycle and observability) — drawn from one pool; every cloud run charges at least platform, Warp-hosted compute adds compute, a Warp model adds AI. The BYOK you configured locally is out of the cloud's reach — cloud runs consume Warp credits as usual; for Enterprise inference in the cloud, the path is BYOLLM, not your local keys. Cloud Runners (compute configs) opened to all users on 2026-07-23 with the oz runner CLI; self-hosted execution stays Enterprise. Oz can also run Claude Code or Codex as cloud harnesses (labeled beta on the pricing page): inference bills your provider, Warp still meters compute and platform — the CLI subscription you already pay for keeps working in the cloud, but not for free.4注 4Note 4Warp Docs · Credits —— 截至 2026-08-06:credits 分三桶:AI(推理)、compute(Warp 托管沙箱,仅 cloud run)、platform(run 生命周期 / integrations / 观测 —— 每个 cloud run 都扣,Business / Enterprise 用自带推理的本地 run 也扣);同一额度池(Settings > Billing and usage);每个 turn 有 credit chip 可 hover 看消耗;个人可以从自己的 credits 经 CLI / API 跑 cloud agents;Slack / Linear integrations 需要团队成员身份;普通 shell 命令不耗 credits。Warp Docs · Credits — as of 2026-08-06: three buckets — AI (inference), compute (Warp-hosted sandbox, cloud runs only), platform (run lifecycle / integrations / observability — charged on every cloud run, and on Business / Enterprise local runs using customer-supplied inference); one shared pool (Settings > Billing and usage); a per-turn credit chip shows usage on hover; individuals can run cloud agents from their own credits via CLI / API; Slack / Linear integrations require team membership; regular shell commands consume no credits.6注 6Note 6Warp Docs · BYOK + changelog 2026-07-23 —— BYOK 不适用于 Cloud Agents:key 只存在你设备上,云端 run 拿不到,cloud runs 照常消耗 Warp credits;旧的「Enterprise team-managed keys」例外已取消,集中托管推理走 BYOLLM。2026-07-23 起 Cloud Runners 与 Cloud Agent Runners 全员开放,配 oz runner CLI;2026-08-06 本机轻测:Warp bundle 带 oz(v0.2026.07.01),oz --help 列出 agent / environment / mcp / schedule / secret 等命令。Warp Docs · BYOK + the 2026-07-23 changelog — BYOK does not apply to Cloud Agents: keys live only on your device, cloud runs can't reach them and consume Warp credits as usual; the old 'Enterprise team-managed keys' exception is gone — centrally managed inference goes through BYOLLM. Since 2026-07-23, Cloud Runners and Cloud Agent Runners are open to all users, with the oz runner CLI; local light check on 2026-08-06: the Warp bundle ships oz (v0.2026.07.01), and oz --help lists agent / environment / mcp / schedule / secret and more.5注 5Note 5Warp Docs · Harnesses in Oz —— Claude Code 与 Codex 可作云端 harness,共享 triggers / environments / secrets / skills / rules / 观测;推理由你提供的凭据记 provider 的账,Warp 计 compute 与 platform credits;在 Cloud Mode 的 harness 下拉、Oz web app 或 API 的 harness 字段选;pricing 页把「Use any harness in the cloud」标为 beta(并列在 Free 下)。Warp Docs · Harnesses in Oz — Claude Code and Codex work as cloud harnesses, sharing triggers / environments / secrets / skills / rules / observability; inference is billed by your provider via credentials you supply, while Warp meters compute and platform credits; pick the harness in Cloud Mode's dropdown, the Oz web app, or the API's harness field; the pricing page labels 'Use any harness in the cloud' as beta (and lists it under Free).验收这一章 —— 主线那两个 run,你要真的派出去、真的验收:Accept the chapter — you really dispatch those two runs, and really review them:
我用 /cloud-agent 或 oz agent run-cloud 派出了两个并行的 cloud run。I dispatched two parallel cloud runs via /cloud-agent or oz agent run-cloud.
报告型任务:artifact 里的结论我抽了两条去源头核对。Report task: I spot-checked two conclusions in the artifact against their sources.
PR 型任务:我读完了 diff,让它过了 CI,不合格就拒。PR task: I read the full diff, let CI run, and rejected what failed.
跑顺的那个改成了 oz schedule create 定时或 /create-skill;一次性的那个停留一次性。The one that worked became an oz schedule create schedule or a /create-skill skill; the one-off stayed one-off.
我知道无人值守的禁止清单:生产写入、删除、付款/发信、deploy、secret 输出。I know the unattended forbidden list: production writes, deletes, payments/email, deploys, secret output.
上云前我确认了 credit 余额 ≥20,知道 BYOK 不覆盖云端。Before going cloud I confirmed a ≥20 credit balance, and I know BYOK doesn't cover the cloud.
在场的活留本地,
离场的活放云端.
Work you attend stays local,
work you leave goes to the cloud.
Aklman Library
— 讨论Discussion
讨论Discussion.
评论区初始化中…Initializing comments…
01 / 01
没有匹配结果No matches.
换个关键词，或按 Esc 回到页面Try another keyword, or press Esc to return
