<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: AgentDrive</title>
    <description>The latest articles on DEV Community by AgentDrive (agentdrive-sh).</description>
    <link>https://dev.to/agentdrive-sh</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Forganization%2Fprofile_image%2F16248%2F42c090c5-7fc5-48cc-95c5-b50a8b76285e.png</url>
      <title>DEV Community: AgentDrive</title>
      <link>https://dev.to/agentdrive-sh</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/agentdrive-sh"/>
    <language>en</language>
    <item>
      <title>Resume your Conductor sessions in any agent</title>
      <dc:creator>AgentDrive</dc:creator>
      <pubDate>Wed, 18 Nov 2026 19:24:26 +0000</pubDate>
      <link>https://dev.to/agentdrive-sh/resume-your-conductor-sessions-in-any-agent-3l68</link>
      <guid>https://dev.to/agentdrive-sh/resume-your-conductor-sessions-in-any-agent-3l68</guid>
      <description>&lt;p&gt;Conductor runs Claude Code and Codex in parallel workspaces. Each session holds the plan, the dead ends and the reasons a change was made, but only the agent that ran it can read that history. AgentDrive syncs those sessions so any coding agent can pick one up: start a task with Claude Code in a Conductor workspace, then continue it in Codex, Cursor, OpenCode or Antigravity, inside Conductor or outside it.&lt;/p&gt;

&lt;h2&gt;
  
  
  1. Install AgentDrive
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;curl -fsSL agentdrive.sh/install &lt;span class="o"&gt;|&lt;/span&gt; sh
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;This installs the &lt;code&gt;adrive&lt;/code&gt; CLI, signs you in, and wires session sync into Claude Code, Codex, Cursor, OpenCode and Antigravity on this machine. From then on, new sessions sync in the background.&lt;/p&gt;

&lt;h2&gt;
  
  
  2. Sync your existing history
&lt;/h2&gt;

&lt;p&gt;Conductor's workspaces run Claude Code and Codex, which &lt;a href="https://www.conductor.build/docs/faq" rel="noopener noreferrer"&gt;keep their files in &lt;code&gt;~/.claude&lt;/code&gt; and &lt;code&gt;~/.codex&lt;/code&gt;&lt;/a&gt;. To upload the sessions you already have, run:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;adrive sync --all
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;It reads sessions oldest first, up to 50 per run. Run it again until it reports &lt;code&gt;Synced 0&lt;/code&gt;. Add &lt;code&gt;--dry-run&lt;/code&gt; to preview the next batch without uploading it. Every plan, Free included, keeps your whole session history.&lt;/p&gt;

&lt;p&gt;Sessions are filed by git remote, so work from a Conductor workspace lands in the same project as the rest of that repo. Sync before you delete a workspace folder: a session whose folder is already gone still uploads, but it goes to a separate project named after the folder.&lt;/p&gt;

&lt;p&gt;Transcripts are redacted for common secret shapes before upload. Turn on Private Workspace with &lt;code&gt;adrive privacy --private-workspace on&lt;/code&gt; if you want them encrypted on your machine first.&lt;/p&gt;

&lt;h2&gt;
  
  
  3. Resume in any agent
&lt;/h2&gt;

&lt;p&gt;Open any wired agent in the repo and ask it to resume your last AgentDrive session. In Claude Code, run &lt;code&gt;/agentdrive:resume&lt;/code&gt;. From a shell:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;adrive resume --list
adrive resume &lt;span class="o"&gt;&amp;lt;&lt;/span&gt;id from the list&lt;span class="o"&gt;&amp;gt;&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The handoff brief carries the goal, the files touched and the most recent turns. It works in any direction: a Codex session resumes in Claude Code just as well as the other way around.&lt;/p&gt;

&lt;h2&gt;
  
  
  What AgentDrive is not
&lt;/h2&gt;

&lt;p&gt;AgentDrive does not run agents in parallel or manage worktrees the way Conductor does. It keeps your sessions and hands work between agents and people. Keep using Conductor to run agents side by side, and use AgentDrive to carry a session from one agent to the next.&lt;/p&gt;

&lt;p&gt;Free includes 10 cloud resumes a month. The &lt;a href="https://agentdrive.sh/pricing" rel="noopener noreferrer"&gt;pricing page&lt;/a&gt; lists the limits on each plan.&lt;/p&gt;

&lt;p&gt;If a Conductor session doesn't come across, email &lt;a href="mailto:hello@agentdrive.sh"&gt;hello@agentdrive.sh&lt;/a&gt; and I'll look at it.&lt;/p&gt;

</description>
      <category>ai</category>
      <category>devtools</category>
      <category>productivity</category>
      <category>claude</category>
    </item>
    <item>
      <title>How background session sync works</title>
      <dc:creator>AgentDrive</dc:creator>
      <pubDate>Wed, 18 Nov 2026 19:18:49 +0000</pubDate>
      <link>https://dev.to/agentdrive-sh/how-background-session-sync-works-4g0l</link>
      <guid>https://dev.to/agentdrive-sh/how-background-session-sync-works-4g0l</guid>
      <description>&lt;p&gt;AgentDrive lets one coding agent resume a session another one started. To make that work without asking you to save anything, it has to sync sessions in the background, safely, and without slowing the agent down. This post walks through how that works in the &lt;code&gt;adrive&lt;/code&gt; CLI and the MCP server. All of the code described here is in the Apache-2.0 client packages.&lt;/p&gt;

&lt;h2&gt;
  
  
  1. A hook in every agent
&lt;/h2&gt;

&lt;p&gt;Each supported agent CLI has a lifecycle hook system, and AgentDrive registers the same command in each one. Antigravity's hook payloads carry no event name, so its entries add &lt;code&gt;--event &amp;lt;name&amp;gt;&lt;/code&gt;.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="nb"&gt;command&lt;/span&gt; -v adrive &lt;span class="o"&gt;&amp;gt;&lt;/span&gt;/dev/null 2&lt;span class="o"&gt;&amp;gt;&amp;amp;&lt;/span&gt;1 &lt;span class="o"&gt;&amp;amp;&amp;amp;&lt;/span&gt; adrive hook &lt;span class="o"&gt;&amp;lt;&lt;/span&gt;harness&lt;span class="o"&gt;&amp;gt;&lt;/span&gt; &lt;span class="o"&gt;||&lt;/span&gt; npx -y @adrive/cli hook &lt;span class="o"&gt;&amp;lt;&lt;/span&gt;harness&lt;span class="o"&gt;&amp;gt;&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The &lt;code&gt;npx&lt;/code&gt; fallback exists because agents launched from a GUI often don't have the global npm bin directory on their &lt;code&gt;PATH&lt;/code&gt;.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Agent&lt;/th&gt;
&lt;th&gt;Where the hook lives&lt;/th&gt;
&lt;th&gt;Events&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Claude Code&lt;/td&gt;
&lt;td&gt;the AgentDrive plugin&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;SessionStart&lt;/code&gt;, &lt;code&gt;Stop&lt;/code&gt;, &lt;code&gt;SessionEnd&lt;/code&gt;, &lt;code&gt;PreCompact&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Codex&lt;/td&gt;
&lt;td&gt;&lt;code&gt;~/.codex/hooks.json&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;SessionStart&lt;/code&gt;, &lt;code&gt;Stop&lt;/code&gt;, &lt;code&gt;SessionEnd&lt;/code&gt;, &lt;code&gt;PreCompact&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Cursor&lt;/td&gt;
&lt;td&gt;&lt;code&gt;~/.cursor/hooks.json&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;sessionStart&lt;/code&gt;, &lt;code&gt;stop&lt;/code&gt;, &lt;code&gt;sessionEnd&lt;/code&gt;, &lt;code&gt;preCompact&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Antigravity&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;~/.gemini/config/hooks.json&lt;/code&gt;, plus &lt;code&gt;~/.gemini/settings.json&lt;/code&gt;, which setup creates or updates&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;SessionStart&lt;/code&gt;, &lt;code&gt;Stop&lt;/code&gt; (in &lt;code&gt;settings.json&lt;/code&gt;: &lt;code&gt;SessionStart&lt;/code&gt;, &lt;code&gt;AfterAgent&lt;/code&gt;, &lt;code&gt;SessionEnd&lt;/code&gt;, &lt;code&gt;PreCompress&lt;/code&gt;)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;OpenCode&lt;/td&gt;
&lt;td&gt;a plugin in &lt;code&gt;~/.config/opencode/plugins/&lt;/code&gt;
&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;session.created&lt;/code&gt;, &lt;code&gt;session.idle&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;&lt;code&gt;adrive init&lt;/code&gt; merges these entries into the existing config files. It keeps any hooks you already have, in their original order, and re-running it produces identical output. Codex only runs hooks it trusts, so setup also records that approval for you; on an older Codex you approve them once via &lt;code&gt;/hooks&lt;/code&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  2. The hook returns fast, and sync runs detached
&lt;/h2&gt;

&lt;p&gt;Hooks run inline with the agent. Codex caps hook timeouts at 3 seconds, so &lt;code&gt;adrive hook&lt;/code&gt; does almost nothing itself. It reads the event payload (session id, working directory, transcript path), records which directory the session belongs to, starts a detached &lt;code&gt;adrive sync&lt;/code&gt; process and exits 0. A hook failure is swallowed, so it can't block or break your session.&lt;/p&gt;

&lt;p&gt;On a session-start event, if auto-context is on, the hook also asks AgentDrive for this project's recent sessions. It puts the newest session from each other agent first and returns up to three as extra context for the agent, with a hint to call &lt;code&gt;drive_resume&lt;/code&gt;. OpenCode's plugin API has no way to inject context, so OpenCode skips this step.&lt;/p&gt;

&lt;p&gt;The detached sync then does the real work:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Locking.&lt;/strong&gt; One sync runs at a time, coordinated through a lock file in &lt;code&gt;~/.config/agentdrive/&lt;/code&gt;. A sync started by a burst of hook events waits up to 30 seconds for the lock instead of giving up.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Debounce.&lt;/strong&gt; A turn-end event (&lt;code&gt;Stop&lt;/code&gt;, &lt;code&gt;stop&lt;/code&gt;, &lt;code&gt;AfterAgent&lt;/code&gt;) re-syncs a session at most every two minutes. Session-end and compaction events skip the debounce, so the final state always lands. OpenCode has no session-end event, so its &lt;code&gt;session.idle&lt;/code&gt; skips the debounce too.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Sweep.&lt;/strong&gt; Each sync also checks every agent for sessions that changed since the last sweep. The first sweep looks back 14 days and handles at most 50 sessions per run. A session that was killed before its end event gets finalized once it has been idle for 30 minutes.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Reading transcripts.&lt;/strong&gt; Each agent stores sessions differently. Claude Code keeps JSONL under &lt;code&gt;~/.claude/projects&lt;/code&gt;, Codex keeps rollout files under &lt;code&gt;$CODEX_HOME/sessions&lt;/code&gt;, Cursor under &lt;code&gt;~/.cursor/projects&lt;/code&gt;, Antigravity under &lt;code&gt;~/.gemini/&amp;lt;app&amp;gt;/brain&lt;/code&gt; and &lt;code&gt;~/.gemini/tmp&lt;/code&gt;, and OpenCode exposes them through &lt;code&gt;opencode session list&lt;/code&gt; and &lt;code&gt;opencode export&lt;/code&gt;. A parser for each one normalizes the session into turns, tool calls and touched paths.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  3. Redaction and ignore rules
&lt;/h2&gt;

&lt;p&gt;Two layers run on your machine before anything is uploaded.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Path rules decide which files can be uploaded at all.&lt;/strong&gt; Artifact sync only looks at files the agent touched in that session. A file is skipped unless every check below passes:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;It is inside your home directory or the session's project.&lt;/li&gt;
&lt;li&gt;It doesn't match the built-in deny list. That list covers &lt;code&gt;.env&lt;/code&gt; and &lt;code&gt;.env.*&lt;/code&gt;, &lt;code&gt;.dev.vars&lt;/code&gt;, &lt;code&gt;.ssh/&lt;/code&gt;, &lt;code&gt;.aws/&lt;/code&gt;, &lt;code&gt;.kube/&lt;/code&gt;, &lt;code&gt;.npmrc&lt;/code&gt;, &lt;code&gt;.netrc&lt;/code&gt;, &lt;code&gt;*.pem&lt;/code&gt;, &lt;code&gt;*.key&lt;/code&gt;, &lt;code&gt;credentials.json&lt;/code&gt;, &lt;code&gt;auth.json&lt;/code&gt;, shell history, SQLite databases, &lt;code&gt;node_modules/&lt;/code&gt;, &lt;code&gt;dist/&lt;/code&gt; and more. The deny list is a hard gate: a project's &lt;code&gt;.adriveignore&lt;/code&gt; can't override it.&lt;/li&gt;
&lt;li&gt;It doesn't match the project's &lt;code&gt;.adriveignore&lt;/code&gt;. This file uses a subset of gitignore syntax: comments, &lt;code&gt;!&lt;/code&gt; negation, &lt;code&gt;/&lt;/code&gt; anchoring, &lt;code&gt;*&lt;/code&gt; and &lt;code&gt;**&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;It is untracked and not ignored by git, or it sits outside any repository. Tracked files are already in git, and gitignored files are where secrets tend to live. If &lt;code&gt;git status&lt;/code&gt; fails for any other reason, the file is skipped.&lt;/li&gt;
&lt;li&gt;It is no larger than 1 MB and doesn't look binary (no NUL byte in the first 8 KB).&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;strong&gt;Content redaction scrubs what does get uploaded.&lt;/strong&gt; The transcript and every synced file go through a pattern-based redactor. It replaces AWS access keys and secrets, GCP service-account private keys, Stripe, GitHub, Anthropic, OpenAI, Slack and AgentDrive tokens, JWTs, PEM private-key blocks, and &lt;code&gt;SOMETHING_KEY=...&lt;/code&gt;-style assignments with markers like &lt;code&gt;[redacted:github-token]&lt;/code&gt;. The version message records which kinds were redacted, never the values.&lt;/p&gt;

&lt;p&gt;The redactor is best-effort, and it is &lt;strong&gt;not a security boundary&lt;/strong&gt;. Pattern matching will miss secrets that don't look like secrets. That's why the path rules come first, and why Private Workspace exists.&lt;/p&gt;

&lt;h2&gt;
  
  
  4. Storage layout
&lt;/h2&gt;

&lt;p&gt;Each session is rendered as plain markdown: a header with the project, working directory, title and turn count, a "Files touched" list, then one &lt;code&gt;## user&lt;/code&gt; or &lt;code&gt;## assistant&lt;/code&gt; section per turn with &lt;code&gt;&amp;gt; tool:&lt;/code&gt; lines for tool calls. The sync writes it to:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;sessions/&amp;lt;project&amp;gt;/&amp;lt;harness&amp;gt;/&amp;lt;session-id&amp;gt;.md
sessions/&amp;lt;project&amp;gt;/files/&amp;lt;path relative to the project&amp;gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;A file outside the project, but inside your home directory, goes under &lt;code&gt;files/_external/&lt;/code&gt; with a short hash of its path in the name.&lt;/p&gt;

&lt;p&gt;The project key is derived from where you work, so every agent files the same repo under the same key:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;With a git remote, it's the normalized remote URL, such as &lt;code&gt;github.com__acme__api&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;With a local repo and no remote, it's the repo folder name plus a short hash of its path.&lt;/li&gt;
&lt;li&gt;Outside a repo, it's &lt;code&gt;local__&amp;lt;folder&amp;gt;-&amp;lt;hash&amp;gt;&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;Sessions from a Claude Code worktree that has since been removed file under its parent repo.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Workspace files are versioned and content-addressed, so re-uploading an unchanged file costs nothing. Each session is also recorded as a run with the id &lt;code&gt;&amp;lt;harness&amp;gt;:&amp;lt;session-id&amp;gt;&lt;/code&gt;. The run carries the session title, its status (&lt;code&gt;in_progress&lt;/code&gt; or &lt;code&gt;success&lt;/code&gt;), the ids of the files it uploaded, the transcript path, the paths it touched and the time of the last turn. Every request is stamped with the agent as its agent platform, which is where provenance comes from: the timeline shows that Codex wrote this file and Claude Code read it.&lt;/p&gt;

&lt;h2&gt;
  
  
  5. Building the handoff brief
&lt;/h2&gt;

&lt;p&gt;When an agent calls &lt;code&gt;drive_resume&lt;/code&gt; with no arguments, AgentDrive picks the latest session in this project from a &lt;strong&gt;different&lt;/strong&gt; agent. Resuming in Codex therefore gives you the Claude Code session, not Codex's own last run. If no other agent has a session there, it falls back to the latest one. You can also pass a specific id from &lt;code&gt;drive_sessions&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;The full transcript would usually be too large to load, so the brief is built to a character budget (12,000 by default, from 1,000 up to 50,000):&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;A one-line intro naming the source agent and session, with an instruction to check file state on disk before relying on it.&lt;/li&gt;
&lt;li&gt;The session metadata.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Goal:&lt;/strong&gt; the first user message, capped at 1,500 characters, or an eighth of the budget if that is smaller.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Files touched&lt;/strong&gt;, plus the file ids of the uploaded copies, so an agent on another machine can fetch them with &lt;code&gt;drive_read&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Recent turns&lt;/strong&gt;, picked from the newest back until the budget runs out and shown in order, with a marker for how many earlier turns were left out.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;If the head is too large, the file lists shrink first, so even a small budget carries the latest turns. Resume works the same way through &lt;code&gt;/agentdrive:resume&lt;/code&gt; in Claude Code and &lt;code&gt;adrive resume&lt;/code&gt; in a shell.&lt;/p&gt;

&lt;h2&gt;
  
  
  6. Private Workspace
&lt;/h2&gt;

&lt;p&gt;Turn it on with &lt;code&gt;adrive privacy --private-workspace on&lt;/code&gt;, and the sync changes in four ways:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Transcripts and synced files are encrypted on your machine with AES-256-GCM before upload. The data key is generated locally and stored wrapped in your AgentDrive config directory. The server only ever sees ciphertext.&lt;/li&gt;
&lt;li&gt;The project key is replaced with a hash (&lt;code&gt;p-&lt;/code&gt; plus 16 hex characters), and synced files are stored under hashed names, so repo names and file paths don't appear in storage paths.&lt;/li&gt;
&lt;li&gt;Run metadata, which the server has to read, leaves out the working directory, the title and the touched paths. The session appears as "Private session".&lt;/li&gt;
&lt;li&gt;The MCP server can't decrypt, so &lt;code&gt;drive_resume&lt;/code&gt; tells the agent to run &lt;code&gt;adrive resume &amp;lt;id&amp;gt;&lt;/code&gt; in a shell. That command decrypts locally and prints the brief.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The trade-off is that server-side features can't read encrypted content. That's the point.&lt;/p&gt;

&lt;h2&gt;
  
  
  Try it
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;curl -fsSL agentdrive.sh/install &lt;span class="o"&gt;|&lt;/span&gt; sh
adrive sync --dry-run    &lt;span class="c"&gt;# parse local sessions and report, upload nothing&lt;/span&gt;
adrive privacy --show    &lt;span class="c"&gt;# see what's on&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The &lt;a href="https://agentdrive.sh/docs/concepts/sessions" rel="noopener noreferrer"&gt;sessions docs&lt;/a&gt; cover the same ground in reference form.&lt;/p&gt;

</description>
      <category>ai</category>
      <category>devtools</category>
      <category>architecture</category>
      <category>privacy</category>
    </item>
    <item>
      <title>Start in any agent, finish in any other</title>
      <dc:creator>AgentDrive</dc:creator>
      <pubDate>Wed, 18 Nov 2026 19:17:38 +0000</pubDate>
      <link>https://dev.to/agentdrive-sh/start-in-any-agent-finish-in-any-other-5h72</link>
      <guid>https://dev.to/agentdrive-sh/start-in-any-agent-finish-in-any-other-5h72</guid>
      <description>&lt;p&gt;I use Claude Code, Codex, OpenCode, Cursor and Antigravity. Most days I use at least three of them. Each one is good at something the others aren't, and every time I switch I end up doing the same chore: copying the plan out of one session, pasting it into the next, explaining what was already tried, and pointing at the files that changed.&lt;/p&gt;

&lt;p&gt;So I was the copy-paste bus between my agents. AgentDrive is what I built so I could stop doing that.&lt;/p&gt;

&lt;h2&gt;
  
  
  What it does
&lt;/h2&gt;

&lt;p&gt;AgentDrive is a shared workspace that lets Claude Code, Codex, OpenCode, Cursor and Antigravity pick up each other's sessions, with provenance on every artifact and portable storage.&lt;/p&gt;

&lt;p&gt;In practice:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Sync runs in the background.&lt;/strong&gt; A hook in each agent CLI hands the session to the &lt;code&gt;adrive&lt;/code&gt; CLI, which uploads a redacted transcript, plus any files the agent wrote that git doesn't track, to your workspace. You don't change how you work.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Any agent can resume.&lt;/strong&gt; Open a different agent in the same project and ask it to continue. It calls &lt;code&gt;drive_sessions&lt;/code&gt; to see recent sessions from every agent and &lt;code&gt;drive_resume&lt;/code&gt; to load a compact handoff brief: the goal, the files touched and the most recent turns. In Claude Code that's &lt;code&gt;/agentdrive:resume&lt;/code&gt;. From a shell it's &lt;code&gt;adrive resume&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Every artifact has provenance.&lt;/strong&gt; Each upload and each session run keeps which agent CLI produced it, so the workspace timeline shows which agent did what, and when.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Session sync is the part you notice first. Underneath it is the same workspace any MCP client can use: 22 MCP tools (18 stable, 4 experimental sandbox tools) for versioned files, semantic search, run logs and share links.&lt;/p&gt;

&lt;h2&gt;
  
  
  Install
&lt;/h2&gt;

&lt;p&gt;One line on macOS or Linux:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;curl -fsSL agentdrive.sh/install &lt;span class="o"&gt;|&lt;/span&gt; sh
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;It installs the &lt;code&gt;adrive&lt;/code&gt; CLI, signs you in through the browser, and wires session sync, the MCP server and the resume and sessions commands into each agent it finds. In Claude Code, AgentDrive is a plugin:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;/plugin marketplace add agentdrive-sh/agentdrive-mcp
/plugin install agentdrive@agentdrive
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Setup also approves the AgentDrive hooks in Codex for you; on an older Codex, approve them once via &lt;code&gt;/hooks&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;Cline, Zed and Continue get the workspace tools as plain MCP clients. They don't have session sync.&lt;/p&gt;

&lt;h2&gt;
  
  
  Privacy
&lt;/h2&gt;

&lt;p&gt;Syncing transcripts means uploading what you and your agents said, so you control all of it with &lt;code&gt;adrive privacy&lt;/code&gt;:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Auto-context:&lt;/strong&gt; at session start, list recent sessions from your other agents in this project.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Artifact sync:&lt;/strong&gt; upload files your agents wrote that git doesn't track, and allow &lt;code&gt;adrive watch&lt;/code&gt; to upload a whole folder (except &lt;code&gt;.gitignore&lt;/code&gt; and &lt;code&gt;.adriveignore&lt;/code&gt; matches).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Transcript sync:&lt;/strong&gt; upload session transcripts.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Private Workspace:&lt;/strong&gt; encrypt transcripts and files on your machine before upload. Synced project and file storage paths are hashed too. Account and session metadata remain managed; direct MCP writes are not encrypted by this CLI setting.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Before anything is uploaded, a redactor scrubs common secret shapes such as cloud keys, API tokens, JWTs and private keys. A built-in deny list skips &lt;code&gt;.env&lt;/code&gt; files, SSH and cloud credentials, and key files, and you can add your own rules in &lt;code&gt;.adriveignore&lt;/code&gt;. Redaction is best-effort, though. It is not a security boundary. Use Private Workspace when synced content must not leave your machine in readable form. If it must not be uploaded at all, turn transcript sync off with &lt;code&gt;adrive privacy --capture-transcripts off&lt;/code&gt; before you start that session.&lt;/p&gt;

&lt;h2&gt;
  
  
  Pricing
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Free, $0:&lt;/strong&gt; the whole loop on one machine. Unlimited local resumes with no account, plus 1 GB storage (soft limit), 1 workspace, 3 agents, and 10 cloud resumes a month.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Pro, $20/month:&lt;/strong&gt; one developer running several agents every day. 50 GB storage (soft limit), 25 agents, and unlimited cross-agent resumes.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Team, $30 per seat per month, 3-seat minimum:&lt;/strong&gt; 3 to 100 seats, 250 GB storage (soft limit), unlimited cross-agent resumes, workspace roles and invites, and bring your own bucket on R2 or S3.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Enterprise:&lt;/strong&gt; custom limits, a security review and a DPA.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Every plan keeps all of your session history. Workspace, seat and agent limits are enforced. Storage is a soft limit: warnings start at 80%, and new files and sessions keep syncing past it. Local resumes on one machine are unlimited and need no account; Free includes 10 cloud resumes a month. Check usage with &lt;code&gt;adrive plan&lt;/code&gt; and upgrade with &lt;code&gt;adrive upgrade&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;Every plan includes session sync and resume across all five agents, Private Workspace, and $0 egress. Encrypted content cannot be searched by meaning on the server, and resume decrypts locally with your key. Details are on the &lt;a href="https://agentdrive.sh/pricing" rel="noopener noreferrer"&gt;pricing page&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  What's next
&lt;/h2&gt;

&lt;p&gt;These are not shipped, so I won't pretend they are: SSO, audit-log export, customer-managed keys, and GCS and Azure buckets. The sandbox tools are experimental. If there's an agent CLI you want synced, or a handoff that didn't carry enough context, tell me. That feedback decides what I build next.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://agentdrive.sh/docs/quick-start" rel="noopener noreferrer"&gt;Get started&lt;/a&gt;, or read &lt;a href="https://agentdrive.sh/docs/concepts/sessions" rel="noopener noreferrer"&gt;how session sync works&lt;/a&gt;.&lt;/p&gt;

</description>
      <category>ai</category>
      <category>devtools</category>
      <category>productivity</category>
      <category>opensource</category>
    </item>
  </channel>
</rss>
