Agent ChannelBot

A TypeScript CLI tool that enables AI agents and humans to interact with Channel Talk workspaces using API credentials (Access Key + Access Secret). Designed for customer support automation, team inbox management, and CI/CD integrations.

Key Concepts

Before diving in, a few things about Channel Talk's terminology:

Channel = workspace (not a chat channel like Slack). The API calls it a "channel," but it means your entire workspace.
UserChats = 1:1 conversations with end users (customers).
Groups = team inbox channels (similar to Slack channels). Referenced by ID or @name.
Managers = human agents on your team.
Bots = automated agents that can send messages and close chats.
Messages use a blocks format: [{ type: "text", value: "..." }]. The CLI handles this automatically when you pass plain text.

Quick Start

# Set your API credentials
agent-channelbot auth set your-access-key your-access-secret

# Verify authentication
agent-channelbot auth status

# Get a workspace overview
agent-channelbot snapshot --pretty

# Send a message to a UserChat
agent-channelbot message send abc123-chat-id "Hello from the CLI!"

# Send a message to a Group
agent-channelbot message send @support "Team update: deployment complete"

Authentication

API Credential Setup

agent-channelbot uses Access Key + Access Secret pairs from Channel Talk's developer settings:

# Set credentials (validates against Channel Talk API before saving)
agent-channelbot auth set your-access-key your-access-secret

# Check auth status
agent-channelbot auth status

# Clear stored credentials
agent-channelbot auth clear

For credential setup, multi-workspace management, and security details, see references/authentication.md.

Memory

The agent maintains a ~/.config/agent-messenger/MEMORY.md file as persistent memory across sessions. This is agent-managed, the CLI does not read or write this file. Use the Read and Write tools to manage your memory file.

Reading Memory

At the start of every task, read ~/.config/agent-messenger/MEMORY.md using the Read tool to load any previously discovered workspace IDs, group IDs, chat IDs, manager IDs, bot names, and preferences.

If the file doesn't exist yet, that's fine. Proceed without it and create it when you first have useful information to store.
If the file can't be read (permissions, missing directory), proceed without memory. Don't error out.

Writing Memory

After discovering useful information, update ~/.config/agent-messenger/MEMORY.md using the Write tool. Write triggers include:

After discovering workspace IDs and names (from auth list, snapshot, etc.)
After discovering group IDs and names (from group list, snapshot, etc.)
After discovering chat IDs (from chat list, etc.)
After discovering manager IDs and names (from manager list, etc.)
After discovering bot names (from bot list, etc.)
After the user gives you an alias or preference ("call this the support workspace", "my main group is X")
After setting a default bot name (from auth bot)

When writing, include the complete file content. The Write tool overwrites the entire file.

What to Store

Workspace IDs with names
Group IDs with names (and @name aliases)
UserChat IDs with context
Manager IDs with names
Bot names and their purposes
Default bot name per workspace
User-given aliases ("support workspace", "billing group")
Any user preference expressed during interaction

What NOT to Store

Never store access keys, access secrets, or any credentials. Never store full message content (just IDs and context). Never store personal user data.

Handling Stale Data

If a memorized ID returns an error (chat not found, group not found), remove it from MEMORY.md. Don't blindly trust memorized data. Verify when something seems off. Prefer re-listing over using a memorized ID that might be stale.

Format / Example

# Agent Messenger Memory

## Channel Talk Workspaces

- `abc123` - Acme Support

## Default Bot (Acme Support)

- Support Bot

## Groups (Acme Support)

- `grp_111` - @support (Support Inbox)
- `grp_222` - @billing (Billing Inbox)
- `grp_333` - @engineering (Engineering)

## Recent UserChats (Acme Support)

- `uc_aaa` - John Doe inquiry (opened)
- `uc_bbb` - Refund request (closed)

## Managers (Acme Support)

- `mgr_001` - Alice (Team Lead)
- `mgr_002` - Bob (Support Agent)

## Bots (Acme Support)

- Support Bot (default, used for auto-replies)
- Notification Bot (used for alerts)

## Aliases

- "support" -> `grp_111` (@support in Acme Support)

## Notes

- Support Bot is used for closing chats and auto-replies
- Notification Bot is used for CI/CD alerts to @engineering

Memory lets you skip repeated group list and chat list calls. When you already know an ID from a previous session, use it directly.

Commands

Auth Commands

# Set workspace credentials (validates against API)
agent-channelbot auth set <access-key> <access-secret>

# Check auth status
agent-channelbot auth status

# Clear all credentials
agent-channelbot auth clear

# List stored workspaces
agent-channelbot auth list

# Switch active workspace
agent-channelbot auth use <workspace-id>

# Remove a stored workspace
agent-channelbot auth remove <workspace-id>

# Set default bot name for sending messages
agent-channelbot auth bot <name>

Message Commands

# Send a message to a UserChat or Group
agent-channelbot message send <target> <text>
agent-channelbot message send abc123-chat-id "Hello!"
agent-channelbot message send @support "Team update"

# List messages from a UserChat or Group
agent-channelbot message list <target>
agent-channelbot message list abc123-chat-id --limit 50

# Get a specific message by ID
# Note: Searches the latest 100 messages. Older messages may not be found.
agent-channelbot message get <target> <message-id>

# Search messages across UserChats
agent-channelbot message grep <query>
agent-channelbot message grep "리뷰" --state opened
agent-channelbot message grep "refund" --chat-limit 100 --limit 50

Search scans UserChats, fetches their messages, and filters by text match (case-insensitive). Options:

Option	Description	Default
`--state <state>`	Filter chats: `opened`, `closed`, `snoozed`, `all`	`all`
`--chat-limit <n>`	Max number of chats to scan	`50`
`--limit <n>`	Max number of results to return	`20`

Target auto-detection: if the target starts with @, it's treated as a group. Otherwise, it's treated as a UserChat. You can override with --type userchat or --type group.

Chat Commands (UserChats)

# List UserChats (default: opened)
agent-channelbot chat list
agent-channelbot chat list --state opened
agent-channelbot chat list --state snoozed
agent-channelbot chat list --state closed

# Get a specific UserChat
agent-channelbot chat get <chat-id>

# Close a UserChat (requires --bot or default bot)
agent-channelbot chat close <chat-id>
agent-channelbot chat close <chat-id> --bot "Support Bot"

# Delete a UserChat (requires --force)
agent-channelbot chat delete <chat-id> --force

Group Commands

# List groups
agent-channelbot group list

# Get a group by ID or @name
agent-channelbot group get <group>
agent-channelbot group get @support

# Get messages from a group
agent-channelbot group messages <group>
agent-channelbot group messages @support --limit 50

Manager Commands

# List all managers
agent-channelbot manager list

# Get a specific manager
agent-channelbot manager get <manager-id>

Bot Commands

# List all bots
agent-channelbot bot list

# Create a new bot
agent-channelbot bot create <name>
agent-channelbot bot create "Deploy Bot" --color "#FF5733" --avatar-url "https://example.com/avatar.png"

# Delete a bot (requires --force)
agent-channelbot bot delete <bot-id> --force

Snapshot Command

Get comprehensive workspace state for AI agents:

# Full snapshot of current workspace
agent-channelbot snapshot

# Filtered snapshots
agent-channelbot snapshot --groups-only
agent-channelbot snapshot --chats-only

# Limit messages per group/chat
agent-channelbot snapshot --limit 10

Returns JSON with:

Workspace metadata (id, name, homepage_url, description)
Groups with recent messages (id, name, messages)
UserChat summary (opened/snoozed/closed counts, recent opened with last message)
Managers (id, name, description)
Bots (id, name)

Output Format

JSON (Default)

All commands output JSON by default for AI consumption:

{
  "id": "msg_abc123",
  "chat_id": "uc_def456",
  "person_type": "bot",
  "plain_text": "Hello world",
  "created_at": 1705312200000
}

Pretty (Human-Readable)

Use --pretty flag for formatted output:

agent-channelbot group list --pretty

Global Options

Option	Description
`--pretty`	Human-readable output instead of JSON
`--workspace <id>`	Use a specific workspace for this command
`--bot <name>`	Use a specific bot name for this command

Common Patterns

See references/common-patterns.md for typical AI agent workflows.

Templates

See templates/ directory for runnable examples:

post-message.sh - Send messages with error handling
monitor-chat.sh - Poll for new UserChats
workspace-summary.sh - Generate workspace summary

Error Handling

All commands return consistent error format:

{
  "error": "No credentials. Run \"auth set <access-key> <access-secret>\" first."
}

Common errors: No credentials, Workspace not found, Bot name is required, Use --force to confirm deletion.

Configuration

Credentials stored in ~/.config/agent-messenger/channelbot-credentials.json (0600 permissions). See references/authentication.md for format and security details.

Config format:

{
  "current": { "workspace_id": "abc123" },
  "workspaces": {
    "abc123": {
      "workspace_id": "abc123",
      "workspace_name": "My Workspace",
      "access_key": "...",
      "access_secret": "..."
    }
  },
  "default_bot": "Support Bot"
}

Limitations

No real-time events / WebSocket connection
No file upload support
No message editing or deletion (Channel Talk API limitation)
No user management (users are end-customers, managed by Channel Talk)
No webhook support
Message search is client-side (no server-side search API) — scans chats sequentially, may be slow with many chats
Plain text messages only (no rich blocks in v1)
Bot name must exist in the workspace for sending messages
Channel Talk has rate limits on API calls

Troubleshooting

`agent-channelbot: command not found`

agent-channelbot is NOT the npm package name. The npm package is agent-messenger.

If the package is installed globally, use agent-channelbot directly:

agent-channelbot message send abc123-chat-id "Hello"

If the package is NOT installed, run it directly with npx -y:

npx -y agent-messenger channelbot message send abc123-chat-id "Hello"

Note: If the user prefers a different package runner (e.g., bunx, pnpx, pnpm dlx), use that instead.

NEVER run npx agent-channelbot, bunx agent-channelbot, or pnpm dlx agent-channelbot. It will fail or install a wrong package since agent-channelbot is not the npm package name.

How to get API credentials

Log in to Channel Talk
Go to Settings > Developers > Open API
Create or copy your Access Key and Access Secret
Run agent-channelbot auth set <access-key> <access-secret>

Rate limiting

Channel Talk enforces rate limits on API calls. The CLI automatically retries on rate limit (429) responses using the Retry-After header. For bulk operations, add delays between requests to avoid hitting limits.

Bot name errors

Some operations (closing chats, sending messages) require a bot name. Set a default with auth bot <name> or pass --bot <name> per command. The bot must exist in your workspace. Use bot list to see available bots.

For other troubleshooting, see references/authentication.md.

agent-channelbot