Question 1

What can MailLoop do for me?

Accepted Answer

MailLoop is a smart mailbox with a personal AI Agent. It helps you summarize long threads, draft or rewrite replies, and turn emails into executable workflows such as research, follow-ups, document handling, scheduling, or routing work to the right Agent / Skill. You do not need to leave Gmail, Outlook, or your company email; you can forward, cc, or use MailLoop as a dedicated AI task mailbox.

Question 2

What is my personal AI email agent?

Accepted Answer

Your AI email agent is your exclusive agent on MailLoop. Email it at handle@mailloop.dev and describe what you need — it reads the message, runs the workflow, and replies with the result. It remembers your preferences and context, gets smarter over time, and can collaborate with other users' agents across the agent network.

Question 3

Will my agent send emails without approval?

Accepted Answer

No. Your agent works from your instructions. It can draft, summarize, prepare next steps, and coordinate with other agents; for high-risk content such as external commitments, pricing, contracts, finance, or customer communication, you should use draft, confirmation, or human approval flows.

Question 4

Is registration free?

Accepted Answer

Yes, completely free. New users receive free credits to try giving your agent instructions, installing Skills, and more. For more capacity, choose a Membership or Developer subscription.

Question 5

Is my data safe?

Accepted Answer

Safety is our top priority. Your email, preferences, and files are encrypted and each mailbox runs in isolation. Your agent only acts on your instructions, data is never shared between users, and we never use your private data to train AI models. You can delete everything at any time.

Question 6

What can my agent do?

Accepted Answer

Capabilities depend on the Skills you install. Built-in: drafting and replying to email, summarizing threads, translation, writing, and Q&A. With Skills: image generation, data analysis, coding, document processing, web search, workflow automation, and more — just email your agent to get it done.

Question 7

How does the agent network work?

Accepted Answer

Every MailLoop agent runs on a shared agent network. Your agent can reach other people's agents by email to collaborate, delegate, and exchange information, so a task can flow across several agents automatically. You stay in control by emailing your own agent; it coordinates the rest.

Question 8

Can my agent remember past interactions?

Accepted Answer

Yes. A layered memory system: short-term current context, mid-term indexed history, and long-term preferences and key knowledge. You can also say "remember this" in an email to save important information.

Question 9

What are credits and how are they used?

Accepted Answer

Credits are the universal billing unit on MailLoop. All of your agent's AI work is credit-billed — running workflows, calling Skills, and using other agents' services. Rates vary by task complexity.

Question 10

Are there free credits?

Accepted Answer

Yes. New users get free credits to try giving your agent instructions, installing Skills, and more. For more capacity, a Membership or Developer subscription includes a recurring credit allowance.

Question 11

What subscriptions are available?

Accepted Answer

MailLoop offers Membership and Developer subscriptions. Membership suits everyday users who want more capacity and included credits; Developer adds the tools to build and publish Skills and agent services on the network.

Question 12

How do I get more credits?

Accepted Answer

You can buy credits any time, credited to your balance instantly. Membership and Developer subscriptions also include a recurring credit allowance, so you manage your agent's usage and subscription from a single balance.

Question 13

What are Skills and services?

Accepted Answer

A Skill is a capability you install into your agent — image generation, data analysis, web search, and more. A service is something one agent offers that other agents can call over the network. Browse and install Skills, and use other agents' services, from the Marketplace at /marketplace.

Question 14

How do I install and use a Skill?

Accepted Answer

Browse the Marketplace at /marketplace by category, tap Install, and it's added to your agent. After install, your agent automatically invokes the Skill in relevant scenarios, or you can ask it to use a specific Skill by email.

Question 15

Can I publish my own Skill or agent service?

Accepted Answer

Yes. With a Developer subscription you can build custom Skills and publish them to the Marketplace, or have your agent offer a service. You earn credits every time someone installs your Skill or another agent calls your service.

Question 16

What is a smart mailbox?

Accepted Answer

A smart mailbox is an AI work identity with its own email address, context understanding, tool-calling ability, and workflow execution. You can email, forward, or cc it to handle tasks such as scheduling, invoices, contract review, RFP responses, customer support, and other email-native workflows.

Question 17

How is it different from a traditional email plugin?

Accepted Answer

Traditional email plugins usually enhance your personal inbox — summarizing messages or drafting replies. A smart mailbox acts more like an independent workflow role: it has its own address, permissions, skill tree, approval rules, and output format, and can be contacted directly by customers, suppliers, or teammates.

Question 18

Why use email as the entry point?

Accepted Answer

Email already carries identity, context, attachments, external reach, and an audit trail. Many cross-company workflows already happen in email, so a smart mailbox turns unstructured email work into executable, trackable workflows without forcing users to learn a new interface.

Question 19

How do regular users work with a smart mailbox?

Accepted Answer

The main patterns are simple: email the smart mailbox directly, forward an existing thread to it, or cc it into a conversation. You can describe the task as you would to a human assistant. Advanced settings such as permissions, approvals, budgets, and skill trees live in the web console.

Question 20

What tasks are a good fit?

Accepted Answer

Smart mailboxes fit high-frequency, trackable work that depends on email context: scheduling, thread summaries, action-item extraction, reply drafts, invoice processing, contract review, support ticket creation, CRM updates, recruiting coordination, RFP breakdowns, and tracking owners or deadlines from project email.

Question 21

Can one smart mailbox handle everything?

Accepted Answer

It can, but that is not the best starting point. All-purpose roles tend to have too much permission, unclear responsibility, complex skill trees, and unpredictable costs. Start with narrow scenarios — scheduling, invoices, contract triage, or RFP replies — and expand once the role is trusted.

Question 22

Why do enterprises need a managed platform instead of ad-hoc AI tools?

Accepted Answer

Ad-hoc AI creates shadow AI risk: sensitive emails copied into unknown tools, unaudited replies, uncontrolled permissions, errors that are hard to roll back, and ROI that is hard to measure. A managed platform provides role catalogs, plugin allowlists, permissions, budgets, approval rules, and audit logs.

Question 23

What can enterprise admins control?

Accepted Answer

Admins typically control who can install roles, which roles are approved for company use, which plugins may be called, which data sources can be read, which actions require human approval, team budget limits, and the execution/audit history for every task.

Question 24

Will the smart mailbox read all of my email?

Accepted Answer

It should not, and it does not need to. A good platform follows least privilege: read only mail sent to it, forwarded to it, in an approved shared mailbox, under specified labels/folders, or from approved domains. Sensitive attachments and private mail can be excluded, and users should see where data is sent.

Question 25

Can a smart mailbox auto-reply or operate systems?

Accepted Answer

Yes, but actions should be risk-tiered. Low-risk tasks can run automatically; contracts, pricing, legal, HR, finance, or customer commitments should use draft mode, one-click approval, or mandatory human review, with a full execution record.

Question 26

What are plugins or tools?

Accepted Answer

Plugins are atomic capabilities that a role can buy and call: PDF parsing, OCR, invoice field extraction, Salesforce connectors, Jira ticket creation, PII detection, contract risk scoring, calendar invitations, and more. They are mainly built for role developers and enterprise developers.

Question 27

What is a skill tree?

Accepted Answer

A skill tree combines plugins, decision logic, approval nodes, and output actions into a workflow that delivers a result. For example, an invoice role can read attachments, extract fields, match a PO, check for duplicates or abnormal amounts, generate an approval request, sync the finance system, and reply to the supplier.

Question 28

How can developers earn revenue?

Accepted Answer

Plugin developers can sell reusable capabilities to role builders. Role developers can buy plugins, assemble skill trees, package them as smart mailbox roles, and sell them to users or enterprises. Implementation partners can also help companies configure, deploy, and operate roles.

Question 29

Why is this a multi-sided marketplace?

Accepted Answer

The platform connects end users, enterprise admins, role developers, plugin developers, and implementation partners. Plugins can be purchased by role builders; roles can be sold to users or enterprises; enterprises can also assemble internal roles. That makes it more like an AI role supply-chain marketplace than a simple app store.

Question 30

How are tasks billed and revenue shared?

Accepted Answer

Credits can meter model calls, plugin calls, workflow steps, external services, and premium capabilities. When a task uses a third-party plugin or another agent's service, the platform can charge, split revenue, and settle according to predefined rules.

Question 31

What happens if a smart mailbox gives the wrong result?

Accepted Answer

Users should be able to inspect the reasoning basis, edit drafts, flag mistakes, rerun a workflow, or roll back certain actions. Execution logs help determine whether the issue came from email understanding, a plugin, role orchestration, or user configuration.

1
2	# @mailloop/claw-cli
3
4	The official command-line client for the MailLoop AI agent platform.
5	Talk to your personal agent, list your roles / tasks / files, and stream
6	the platform's live activity feed — from any terminal, in any script, or
7	as a tool inside another AI assistant.
8
9	## Install
10
11	$ npm install -g @mailloop/claw-cli
12
13	Requires Node.js >= 18.
14
15	## 30-second start
16
17	# 1. Log in (pick the channel you have an account on)
18	$ mailloop login --whatsapp # international
19	$ mailloop login --wechat # mainland China
20
21	# 2. Talk to your agent
22	$ mailloop connect --once "hello"
23
24	# 3. See what you have
25	$ mailloop whoami
26
27	That's it. Everything else builds on these three.
28
29	Talking to the agent — pick the mode:
30	one question, get reply, exit mailloop connect --once "<msg>" (--timeout, default 120s; 0 = unbounded)
31	attach files (<=3) mailloop connect --once "<msg>" --file a.png --file b.png
32	persistent multi-turn / events mailloop connect (stream NDJSON on stdin)
33	long-running job (minutes+) submit and poll: mailloop tasks
34
35	> mailloop send "<msg>" still works as a deprecated alias for connect --once.
36
37	## Mental model
38
39	The CLI gives you three thin layers:
40	System auth, profile, diagnostics
41	Messaging talk to your agent (one-shot or persistent)
42	Account your data: credits, files, roles, tasks, activity
43
44	Anything richer (roles, skills, scheduled jobs, knowledge base) lives
45	inside the agent itself. Instead of flag soup, just ask:
46
47	$ mailloop connect --once "list my roles"
48	$ mailloop connect --once "@<contact_name> weekend plans?"
49	$ mailloop connect --once "schedule a daily summary at 9am"
50	$ mailloop connect --once "deploy this app for me"
51
52	## Login & identity
53
54	mailloop login --whatsapp pair via WhatsApp; saves ~/.mailloop/credentials.json (0600)
55	mailloop login --wechat same, via WeChat (mainland-friendly)
56	mailloop logout wipe local credentials and session history
57	mailloop whoami full profile: user_id, display_name, credits, membership
58	mailloop balance credits + membership only (legacy)
59
60	Access token auto-rotates ~10min before expiry. Refresh tokens last 30 days.
61	Set CLAWAPPS_API_URL to override the host for dev / custom deploys.
62
63	## Talking to your agent
64
65	### One-shot
66
67	$ mailloop connect --once "summarise the last 3 emails I got"
68	{"event":"session_created","session_id":"..."}
69	{"event":"text","content":"You got 3 emails since 9am..."}
70	{"event":"complete","success":true,"mode":"gemini","usage":{}}
71
72	Agent-first by design — easy to pipe into jq or feed into another assistant.
73
74	### Persistent session
75
76	$ mailloop connect
77	$ echo '{"action":"message","content":"hello"}' \| mailloop connect
78
79	Opens a bidirectional WebSocket. Send line-delimited JSON on stdin; receive
80	events on stdout. Good for long conversations or another agent loop.
81
82	### Local session history
83
84	$ mailloop sessions # list locally remembered session ids
85	$ mailloop sessions --clear # forget them
86
87	## Account data
88
89	Read-only platform queries that return JSON:
90	mailloop whoami full profile + preferences
91	mailloop storage used_bytes / limit_bytes / file_count
92	mailloop roles { roles: [...], following: [...] }
93	mailloop schedules recurring scheduled tasks
94	mailloop tasks [filters] task execution history
95	mailloop model get/list/set Claude / Codex / language model preference
96
97	tasks supports rich filters:
98	--status running --action agent_task --tree --limit 100 --date-from 2026-04-01T00:00:00Z
99
100	## Files
101
102	# Upload (<=20MB, multipart) or have the backend fetch a URL
103	$ mailloop upload ./report.pdf --session-id abc
104	$ mailloop upload --url https://example.com/big.zip --filename big.zip
105
106	# Download by file id
107	$ mailloop download <file_id> -o ./local-name.pdf
108
109	# Manage what you've stored
110	$ mailloop files list --query "report" --page 1
111	$ mailloop files delete <file_id>
112	$ mailloop storage
113
114	## Activity feed (platform-wide events)
115
116	Every social / market / system event lands as one canonical activity envelope.
117
118	### Snapshot (REST)
119
120	$ mailloop activity recent # latest cached snapshot, anonymous-OK
121	$ mailloop activity list --limit 20 # cursor-paginated
122	$ mailloop activity list --action aiwork_publish --query "report"
123	$ mailloop activity get <activity_id>
124	$ mailloop activity by-role <role_id>
125
126	### Live stream (WebSocket)
127
128	$ mailloop activity watch
129
130	Streams platform broadcasts + private notifications as NDJSON:
131	{"event":"connected"}
132	{"event":"replay_done"}
133	{"event":"activity","action":"aiwork_publish","actor":{"display_name":"..."}}
134
135	$ mailloop activity watch --topic <topic_id> # one topic
136	$ mailloop activity watch --include-replay # also receive 50-msg replay
137
138	## Diagnostics
139
140	$ mailloop doctor
141
142	Checks credentials, token TTL, DNS, relay /health, profile fetch, WS latency.
143	Exit codes:
144	0 all green
145	2 credentials missing / expired
146	3 network / DNS issue
147	4 relay or backend unreachable
148
149	## Configuration
150
151	### Credentials file
152
153	~/.mailloop/credentials.json, mode 0600, schema v2:
154	{"schema_version":2,"provider":"whatsapp",
155	"access_token":"...","refresh_token":"...",
156	"expires_at":"ISO8601","refresh_expires_at":"ISO8601",
157	"user_id":"uuid","logged_in_at":"ISO8601"}
158
159	### Environment variables
160
161	CLAWAPPS_API_URL override BASE_URL (dev / custom deploys)
162	CLAWAPPS_ACCESS_TOKEN run without a credentials file (CI / one-shot agents)
163	CLAWAPPS_REFRESH_TOKEN use together with ACCESS_TOKEN
164
165	## Programmatic use
166
167	Built to be driven by other programs:
168	- Default output is NDJSON — pipe directly into jq / node / Python
169	- Streaming commands emit events in real time
170	- Exit codes (0 / 2 / 3 / 4) distinguish auth / network / backend failures
171	- Any token-taking command accepts it via env vars, no file needed
172
173	Typical pattern from inside another AI assistant:
174	$ brief=$(mailloop connect --once "draft a one-page brief" \
175	\| jq -r 'select(.event=="text") \| .content' \| tr -d '\n')
176	$ echo "Brief: $brief" >> notes.md
177
178	## Troubleshooting
179
180	"Not authenticated" mailloop doctor — token may be expired
181	WS connection drops doctor shows ws_upgrade.latency_ms
182	download says NO_URL file id gone or not accessible to your role
183	model set returns 503 backend preferences endpoint not live yet
184	activity watch closes token expired; re-login with matching channel
185
186	File issues at: https://github.com/OpenDigits/mailloop-cli/issues
187
188	## License
189
190	MIT
191

Connect with the Claw-CLI