You have spent weeks teaching your agents in another tool: their personalities, their skills, what they know about you. Starting again from a blank page is the main reason people put off trying something new. Bring your setup is there so you do not have to.
It reads what you already set up in OpenClaw, Hermes or OpenBot (and Claude Code, Codex and Cursor too), shows you every item before anything is added, and keeps a record so the whole import can be undone. This guide covers how it works, what comes across from each tool, what stays behind, and how your keys are handled.
How Bring your setup works
Bring your setup looks in a short list of known places under your home folder. Nothing is searched for, and nothing in the other tool is written to: it only reads.
- Open Settings, Data, Bring your setup. The first run of Bloks also offers it as an optional step when it finds something.
- Read the review. Every item has a preview and says where it would go, grouped by what it becomes here.
- Settle any clashes. A name that is already taken is yours to settle. A skill that ships with Bloks is never replaced.
- Fill in or tick the key slots you want (see below).
- Import. Only the items you left ticked come across.
If something turns up in more than one tool’s folder (Cursor, for example, reads Claude Code’s skills), the second copy waits unticked so you do not get it twice.
Importing again later replaces the text it added before, in place, rather than adding a second copy.
From another device, or a server
If your agents live on a bloks-server, you, as the owner, can run Bring your setup from a paired device against that server. The page names the computer it reads from, so you always know whose home folder is being read. Agents and members cannot use it.
What comes across from OpenClaw
From ~/.openclaw/openclaw.json, Bring your setup reads your workspace folder, your default model, and your Telegram, Slack and Discord channels. Then, from the workspace:
SOUL.md,IDENTITY.mdandAGENTS.md, the files that say who your agent is;MEMORY.mdandmemory/*.md, which become its memory and topic files;USER.md, whose lines become suggestions in About you, for you to keep or dismiss;- the skills in the workspace, and the managed skills in
~/.openclaw/skills.
Instructions become the agent’s instructions when they fit, or a block in its MEMORY.md when they do not. WhatsApp, Signal and other channels are named in the review as left out.
What comes across from Hermes
From ~/.hermes, Bring your setup looks for SOUL.md, the memories folder and your skills. Each is looked for rather than assumed, so if your setup is laid out a little differently, the review shows what it found and nothing more.
What comes across from OpenBot
OpenBot keeps several agents, each with its own parts. Bring your setup finds them in ~/OpenBot/Agents and in OpenBot’s app data folder, and brings each agent across with its name, instructions, skills and memories, previewed before anything is added.
- Instructions come from each agent’s instruction file.
- Memories marked essential become a block in the agent’s
MEMORY.md. The rest go to a topic file,memory/From OpenBot.md. - Engines map to the same family here: Codex to Codex, Claude Code to Claude Code, and likewise for Grok, OpenCode, Gemini, Cursor, Cline, Devin, Pi, Kimi, OpenRouter and DeepSeek. An agent on GitHub Copilot, Hermes, Muse, a custom endpoint or a custom ACP agent comes across on your default engine, and says so.
- Tool servers go through the same checks as any other and are attached to no agent until you choose.
The OpenBot transfer file
An OpenBot transfer file brings more: the agents’ pictures, their workspace files and their routines. Pick it in Bring your setup in the app, or upload it from a browser on bloks-server.
- Workspace files arrive unticked. When you tick them, they are copied into the agent’s folder, never over a file already there. Hidden files and anything named like a credential stay out.
- Routines also arrive unticked, and every one you bring is filed paused until you switch it on. A routine whose schedule cannot be read is set to 09:00 every day, and the review says so.
- The file is checked as a whole before anything in it is read, and one that does not read cleanly is refused.
What stays behind
Some things are deliberately left where they are, and the review says so under each source:
- Chat history. Conversations from OpenClaw, Hermes and OpenBot do not come across.
- OpenBot’s channels, saved logins and vault, connectors and browser profiles.
- Channels Bloks cannot carry, such as OpenClaw’s WhatsApp and Signal, are named as left out.
Chat apps that do come across (OpenClaw’s Telegram, Slack or Discord) are connected through the same checks Settings uses, then saved switched off, because the other tool may still be answering on that bot. Turn each on when you have moved over.
Your keys never travel without a yes
Bring your setup treats keys with care:
- No credential file is opened. Files named like one (
auth.json,.credentials.json,.env, tokens, keys, keychains) are never read. - Text is scrubbed. Anything key-shaped in instructions, memory or skills is taken out, and the item says how many were removed.
- Keys are slots. A tool server’s environment and header values, and a chat app’s token, show as fields in the review. Each one is typed by you, or copied from the other tool only because you ticked that one slot.
- Variables stay variables. A value written as a variable, such as
${NAME}, is named as one and never resolved.
Rules are careful in the same way. Each waits for you to tick it, and an allow rule that would let agents do more than before is left out with the reason.
Undo the whole import
Changed your mind? Every import can be undone. Bloks keeps the newest 20 imports, step by step, and walks one back newest first.
- Each step goes back only if what is there is still what the import left.
- Anything you changed since, a suggestion you kept, or an agent you have already talked with stays, and says why.
- Agents the import made are archived, not deleted.
Try it from a terminal first
To see what would come across before you open the app:
bloks-server import --dry-run
For OpenBot with a transfer file:
bloks-server import openbot --dry-run --transfer FILE
Add --json for output a script can read. A dry run needs no server and writes nothing. The import itself happens only in the app, where you answer the slots.
After the move
Once your agents are in, a few things are worth a look:
- Open Memory on the main rail and read what came across. Edit or forget anything that no longer fits. Give an AI agent long-term memory you can read explains how memory works here.
- Check each agent’s engine in its settings, especially any that came across on the default engine.
- Switch on the routines and chat apps you want running, one at a time.
- Pair your phone, as in Control your AI agents from your iPhone.
If you are weighing up the switch, the comparison pages set Bloks beside each tool, and the docs cover the rest.
Questions
Will Bring your setup copy my API keys?
Only the ones you tick. No credential file is opened, text is scrubbed of anything that looks like a key, and each key or token is a slot you type or tick one by one.
Does my chat history come across from OpenClaw, Hermes or OpenBot?
No. Chat history stays where it is, and the review says so. Bloks can bring past Claude Code and Codex sessions over as conversations, from Settings, Past conversations.
Can I undo an import?
Yes. Every import can be undone. Each step goes back only if it is still what the import left, and agents it made are archived, not deleted.
Can I see what would come over before I start?
Yes. bloks-server import --dry-run prints the same scan from a terminal, with nothing written. The import itself happens in the app, where you answer the review.
Will my OpenClaw Telegram bot answer twice?
No. A chat app brought over is saved switched off, because the other tool may still be answering on that bot. Switch it on in Bloks when you are ready.