Which Personal Assistant Preserves the Right Reply Identity?
Test what recipients actually see. Use NanoClaw’s per-agent Slack identities where visible separation matters, and trial OpenClaw’s shared WhatsApp identity where separate workspaces can sit behind one sender.
Test the identity a teammate sees
NanoClaw’s repository describes one Slack app per agent, with separate bot identities, containers and memory. Turn that architecture into a recipient-side test. In a scratch workspace, assign a calendar persona and an expense persona distinct synthetic reference codes. From the same test room, send each persona the same request for its own reference code. Have a second observer record the displayed bot identity and returned code without relying on the operator’s view. Then ask for each persona’s code in the other persona’s conversation. Your pass condition is the intended identity and the intended code for normal requests, plus the expected refusal or missing-record response for the cross-persona requests. If the bot name is right but the code belongs to the other persona, investigate shared room context before assuming that separate containers provide the data boundary you need.
GitHub - nanocoai/nanoclaw: Your own always-on AI assistant: self-hosted, open source, lightweight. Each agent runs in its own container on your machine, on the harness you choose (Claude Code, Codex, Cursor, OpenCode). Talk to it from Slack, WhatsApp, Telegram, Discord and more. · GitHubCheck what survives behind a shared number
OpenClaw documents routing multiple WhatsApp senders into different agent workspaces and session stores behind one replying account. That arrangement can work for a household organizer if everyone expects replies from the same number. It fails a requirement for separate externally visible identities even when workspace selection is correct. Test it with two consenting test senders, each given a different synthetic appointment code. Record the account shown on each handset, the returned code and elapsed seconds. Repeat after restoring the assistant state in an isolated test setup, keeping the production connection out of the exercise. If the correct code appears only after you re-enter it, the exercise demonstrated reconnection and a missing state dependency. For households with shared devices, also test from the actual sender identity the gateway receives, since routing by sender cannot distinguish two people using the same account.
FAQ: providers, hardware, and hosting - OpenClawTest the work before connecting accounts
Claude is documented to produce editable files such as documents, spreadsheets, and slides. Use it as a baseline if you mostly want to turn notes into something you can review. If you already use ChatGPT or Gemini, put each through the same trial before paying for a plan or connecting a live inbox. Verify scheduling, delivery channels, and export in the specific account you would use if any of them is a requirement.
For an illustrative trial, give each candidate the same three fictional meeting notes, one date correction, and a request for a one-page agenda. Pass if the file opens in your editor, uses the corrected date, and flags the missing decision owner. Time the cleanup. A polished draft that carries forward the wrong date still fails.
Claude | The AI assistant for workProve the WhatsApp path end to end
OpenClaw’s WhatsApp integration uses a linked session owned by the Gateway. On a test OpenClaw instance, start channel setup with openclaw channels add --channel whatsapp and follow the documented plugin and pairing flow. Use a test account or conversation you control. Send one message from the phone, close the desktop session you used for setup, and send another. Restart the Gateway during a planned test and check whether delivery resumes without pairing again.
For Hermes, verify the connector for your intended channel, where its linked state lives, and what a restart requires. Then repeat the same delivery and restart checks on that candidate. Record whether either assistant misses a message, replies twice, or needs a phone intervention. For unattended use, require a separate scheduled delivery trial at a specific local time.
WhatsApp - OpenClawRehearse a move with three records
NanoClaw documents separate Slack bot identities, containers, and memory for its agents. If you are choosing it for distinct personal and work routines, start in a test Slack workspace and confirm which identity handles each message. Check the shared-room behavior with harmless fixtures before adding real files. Treat distinct bot identities and recovered assistant state as separate acceptance checks.
Suppose your illustrative test data is a dietary preference, a correction to that preference, and a recurring reminder in a named time zone. Use the candidate’s supported export or backup path and restore into a clean instance or replacement environment. Pass if the corrected preference is recalled and the reminder fires once at the expected time. Record any account pairing that required manual setup and any history you could only read as an archive.
GitHub - nanocoai/nanoclaw: Your own always-on AI assistant: self-hosted, open source, lightweight. Each agent runs in its own container on your machine, on the harness you choose (Claude Code, Codex, Cursor, OpenCode). Talk to it from Slack, WhatsApp, Telegram, Discord and more. · GitHubCount the devices your channel needs
For iMessage, OpenClaw documents using imsg on a Mac signed into Messages. You can put the Gateway on that Mac, or keep the Gateway on Linux and use an SSH wrapper to reach imsg on the Mac. Compare the simpler single-machine setup with the separate Gateway if you already have a Linux host and need it for other parts of the assistant.
Test both paths using an account and recipient you control. On the separate-host setup, take the test Mac offline briefly, send a harmless request, and record what happens when the Mac returns. Check for a visible failure and verify that recovery produces no duplicate replies. If this device dependency is unworkable, trial the WhatsApp path or a reviewed draft workflow and measure the inconvenience against your need for iMessage.
FAQ: providers, hardware, and hosting - OpenClawSources
- GitHub - nanocoai/nanoclaw: Your own always-on AI assistant: self-hosted, open source, lightweight. Each agent runs in its own container on your machine, on the harness you choose (Claude Code, Codex, Cursor, OpenCode). Talk to it from Slack, WhatsApp, Telegram, Discord and more. · GitHub
- FAQ: providers, hardware, and hosting - OpenClaw
- Claude | The AI assistant for work
- WhatsApp - OpenClaw