Config: channels, tools & integrations

This page is the field reference for the parts of ~/.revka/config.toml that decide how the agent talks to the outside world: the [channels_config] namespace that wires up messaging platforms, the web and browser tool sections, the third-party integration sections (Composio, Google Workspace, Microsoft 365, Notion, Jira, LinkedIn), and the media and operational tool sections. Every key lists its type, default, and meaning so you can edit config.toml directly.

Use this page when you are turning channels and tools on or off, scoping their reach, or hardening a deployment. For the conceptual walkthrough of connecting a channel, see Connect a messaging channel; for the tool catalog, see the Tools overview. For provider, agent, and routing keys see Config: provider, agent & routing; for gateway, memory, and security keys see Config: gateway, memory, security & platform.

Caution

Channel allowlists are deny-by-default. Every channel allowlist defaults to [] (deny all). You must explicitly add senders. Most integration tool sections are also gated on enabled = false and only register the tool when set to true.

Web/browser tools default to open. When the [browser], [http_request], [web_fetch], or [web_search] section is absent from your config (or as written by revka onboard), the effective default is enabled = true with allowed_domains = ["*"] (all public hosts; local/private hosts remain blocked). The enabled = false / allowed_domains = [] deny-by-default only applies when you add one of these sections and omit those specific fields.

[channels_config] — messaging channels

[channels_config] holds one top-level timeout plus one [channels_config.<name>] sub-table per channel. Channels are deny-by-default: a channel with no allowlist entries accepts nothing. A single entry "*" allows all.

[channels_config]
message_timeout_secs = 300

[channels_config.telegram]
bot_token = "123456:TELEGRAM-TOKEN"
allowed_users = ["123456789"]   # [] = deny all, "*" = allow all
stream_mode = "off"             # off | partial
mention_only = false
interrupt_on_new_message = false

[channels_config.discord]
bot_token = "DISCORD-BOT-TOKEN"
guild_id = "123456789012345678" # optional
allowed_users = ["*"]
stream_mode = "multi_message"   # off | partial | multi_message

Top-level and common keys

Key	Type	Default	Meaning
message_timeout_secs	u64	300	Base per-message processing budget. Scaled by tool-loop depth up to the cap in [pacing].message_timeout_scale_max.
cli	bool	true	CLI terminal stdin/stdout channel (also revka chat). Enabled by default; set cli = false to disable it.

Most channels share these fields (names vary per platform):

Field	Meaning
bot_token / api_key	Channel credential.
allowed_users / allowed_numbers / allowed_senders / allowed_from / allowed_pubkeys / allowed_contacts / allowed_destinations	Sender allowlist. [] = deny all, "*" = allow all. The exact field name varies — see the table below.
mention_only	Only respond when @mentioned (group chats).
stream_mode	"off", "partial" (live draft edits), or "multi_message" (paragraph splits) where supported.
proxy_url	Per-channel HTTP/SOCKS5 proxy.
notification_chat_id / notification_channel_id	Target for unsolicited notifications.
interrupt_on_new_message	Cancel in-flight generation when the same sender sends again. Supported on Telegram, Slack, Discord, Mattermost, and Matrix.

Allowlist field by channel

The allowlist field name differs by platform. An empty list denies all; "*" allows all.

Channel	Allowlist field
Telegram, Discord, Slack, Mattermost, Matrix, IRC, Lark, Feishu, DingTalk, QQ, Nextcloud Talk, WeCom, Mochat, Twitter	allowed_users
Signal	allowed_from
WhatsApp (Cloud & Web), WATI	allowed_numbers
Email, Linq, Gmail Push	allowed_senders
iMessage	allowed_contacts
Nostr	allowed_pubkeys
ClawdTalk	allowed_destinations

Platform-specific fields

Selected channels need extra keys. See Channels overview and the per-platform pages for full setup runbooks.

WhatsApp

Mode is auto-negotiated from which keys you set: phone_number_id → Cloud API; session_path → Web mode.

# Cloud API (Meta webhook mode — needs a public HTTPS callback)
[channels_config.whatsapp]
access_token    = "EAAB..."
phone_number_id = "123456789012345"
verify_token    = "your-verify-token"
app_secret      = "your-app-secret"     # HMAC payload signature (recommended)
allowed_numbers = ["*"]

# Web mode (wa-rs) — requires the `whatsapp-web` build feature
[channels_config.whatsapp]
session_path    = "~/.revka/whatsapp-session.db"
pair_phone      = "15551234567"          # omit for QR-code flow
mode            = "business"             # or "personal"
allowed_numbers = ["*"]

Slack

[channels_config.slack]
bot_token         = "xoxb-..."
app_token         = "xapp-..."           # app-level token for Socket Mode
channel_ids       = ["C123...", "D456..."] # takes precedence over channel_id
allowed_users     = ["*"]
thread_replies    = true
use_markdown_blocks = true               # 12k-char messages vs 3k
cancel_reaction   = "x"                   # emoji that cancels in-flight work

Matrix

Requires the channel-matrix build feature. E2EE works transparently with the right device_id.

[channels_config.matrix]
homeserver    = "https://matrix.example.com"
access_token  = "syt_..."
user_id       = "@revka:matrix.example.com"
device_id     = "DEVICEID123"
room_id       = "!room:matrix.example.com"
allowed_rooms = []                       # extra rooms to accept
allowed_users = ["*"]
stream_mode   = "partial"
recovery_key  = ""                       # cross-signing recovery

Nostr / Signal

# Nostr — requires the `channel-nostr` build feature; key is a high-value secret
[channels_config.nostr]
private_key     = "nsec1..."
relays          = ["wss://relay.damus.io", "wss://nos.lol"]
allowed_pubkeys = ["hex-or-npub-key"]    # [] = deny all, "*" = allow all

# Signal — bridges to a local signal-cli daemon
[channels_config.signal]
http_url    = "http://127.0.0.1:8686"
account     = "+1234567890"
group_id    = ""                         # "dm" for DM-only
allowed_from = ["*"]

Linq / Nextcloud Talk

# Linq (iMessage / RCS / SMS via Linq Partner API) — webhook mode
[channels_config.linq]
api_token       = "linq-partner-api-token"
from_phone      = "+15551234567"
signing_secret  = "optional-signing-secret"  # env: REVKA_LINQ_SIGNING_SECRET
allowed_senders = ["*"]

# Nextcloud Talk — webhook mode, HMAC-signed
[channels_config.nextcloud_talk]
base_url       = "https://cloud.example.com"
app_token      = "nextcloud-talk-app-token"
bot_name       = "revka"
webhook_secret = "optional-webhook-secret"   # env: REVKA_NEXTCLOUD_TALK_WEBHOOK_SECRET
allowed_users  = ["*"]

Note

Hot-reload: root keys default_provider, default_model, api_key, api_url, and everything under [reliability] are reloaded on the next inbound channel message — no restart. Channel credentials themselves require a restart (or revka onboard --repair-channels).

The [transcription], [tts], [media_pipeline], and [multimodal] sections below are also consumed by channels (voice notes, audio attachments, image handling).

Web & browser tools

These sections gate the agent’s outbound web access. When absent from your config, all four default to enabled = true with allowed_domains = ["*"] (all public domains). See the note in the Aside above.

[browser] — browser automation

Backs the browser and browser_open tools. Supports pluggable backends.

[browser]
enabled = true
allowed_domains = ["example.com"]   # "*" = all public domains (local/private still blocked)
backend = "agent_browser"           # agent_browser | rust_native | computer_use | auto
native_headless = true

[browser.computer_use]
endpoint = "http://127.0.0.1:8787/v1/actions"
allow_remote_endpoint = false

Key	Type	Default	Meaning
enabled	bool	true	Enable the browser tools (effective default when section is absent).
allowed_domains	list	["*"]	Domain allowlist. "*" = all public domains; local/private hosts always blocked. Effective default when section is absent.
backend	string	"agent_browser"	agent_browser, rust_native (needs the browser-native feature), computer_use, or auto.
native_headless	bool	true	Headless mode for the rust-native backend.
native_webdriver_url	string	http://127.0.0.1:9515	WebDriver endpoint for the native backend.
native_chrome_path	string?	unset	Chrome/Chromium binary path.
computer_use.allow_remote_endpoint	bool	false	Guards against exposing the computer-use sidecar publicly.

[browser_delegate] — delegate to a browser CLI

Backs the browser_delegate tool, which hands browser tasks to a CLI subprocess (e.g. Claude Code with the claude-in-chrome MCP). Requires shell access.

[browser_delegate]
enabled = true
cli_binary = "claude"
chrome_profile_dir = "~/.config/chrome-corp-profile"
allowed_domains = ["teams.microsoft.com", "outlook.office.com"]
blocked_domains = []
task_timeout_secs = 120

[http_request] — arbitrary HTTP

Backs the http_request tool (full GET/POST/PUT/DELETE/PATCH/HEAD/OPTIONS).

[http_request]
enabled = true
allowed_domains = ["api.example.com", "api.github.com"]
max_response_size = 1000000
timeout_secs = 30
# allow_private_hosts = false

Key	Default	Meaning
enabled	true	Enable the tool (effective default when section is absent).
allowed_domains	["*"]	Domain allowlist. "*" = any public domain; local/private still blocked unless allow_private_hosts. Effective default when section is absent.
max_response_size	—	Cap response bytes.
timeout_secs	—	Per-request timeout.
allow_private_hosts	false	Permit private/loopback targets.

[web_fetch] — fetch a page as text

Backs the web_fetch tool (GET-only, HTML → readable text, with optional Firecrawl fallback for JS-heavy sites).

[web_fetch]
enabled = true
allowed_domains = ["docs.anthropic.com", "*"]
blocked_domains = []
max_response_size = 1000000
timeout_secs = 30

[web_fetch.firecrawl]
# api_key = "fc-..."        # enables the JS-rendering fallback

Other keys: allowed_private_hosts (per-host exception to the private-host block) and web_fetch.firecrawl.* (Firecrawl fallback). Distinguish from http_request: web_fetch is opinionated GET-only-to-text; http_request is full HTTP.

[web_search] — web search

Backs the web_search_tool. Web search is enabled by default with DuckDuckGo (no API key).

[web_search]
enabled = true
provider = "duckduckgo"     # duckduckgo (free) | brave | searxng
max_results = 5
timeout_secs = 15
# brave_api_key = "..."           # for Brave
# searxng_instance_url = "https://searx.example.com"  # for SearXNG

Key	Default	Meaning
enabled	true	Enable web search.
provider	"duckduckgo"	duckduckgo, brave, or searxng.
max_results	5	Result count.
timeout_secs	15	Search timeout.
brave_api_key	unset	Required for the Brave provider (env BRAVE_API_KEY).
searxng_instance_url	unset	Required for self-hosted SearXNG (env SEARXNG_INSTANCE_URL).

Tip

These settings also have env equivalents: WEB_SEARCH_ENABLED, WEB_SEARCH_PROVIDER, WEB_SEARCH_MAX_RESULTS, WEB_SEARCH_TIMEOUT_SECS.

Integration sections

Each integration section gates one tool and supplies its credentials. All are enabled = false by default. Several read credentials from env vars as well as config.

[composio] — managed OAuth (1000+ apps)

Backs the composio tool. Requires a Composio API key (config api_key or env COMPOSIO_API_KEY).

[composio]
enabled = true
api_key = "comp_..."
entity_id = "default"

Connect an app’s OAuth via composio connect. If Composio returns a missing connected-account error, call the tool’s list_accounts action and pass the returned connected_account_id to execute.

[google_workspace] — Gmail, Drive, Calendar, etc.

Backs the google_workspace tool via the gws CLI. Requires gws installed (npm install -g @googleworkspace/cli) and authenticated, plus shell access.

[google_workspace]
enabled = true
default_account = "[email protected]"
allowed_services = ["gmail", "drive"]
audit_log = true

[[google_workspace.allowed_operations]]
service = "gmail"
resource = "users"
sub_resource = "messages"   # Gmail needs the 4th segment
methods = ["list", "get"]

Key	Meaning
enabled	Enable the tool.
default_account / credentials_path	Account selection and credentials.
allowed_services	Service allowlist (gmail, drive, calendar, sheets, docs, …).
allowed_operations[]	Fine-grained service/resource/sub_resource/methods scoping. Gmail commands are 4-segment (include sub_resource); Drive/Calendar are 3-segment (omit it).
rate_limit_per_minute / timeout_secs	Throttle and timeout.
audit_log	Log each operation.

[microsoft365] — Outlook, Teams, Calendar, OneDrive, SharePoint

Backs the microsoft365 tool via Microsoft Graph. Requires an Azure AD app registration.

[microsoft365]
enabled = true
tenant_id = "..."
client_id = "..."
client_secret = "..."        # required for client_credentials flow
auth_flow = "client_credentials"   # or device_code
scopes = ["Mail.Read", "Chat.ReadWrite"]
token_cache_encrypted = true
# user_id = "..."            # target user for app-only flows

The token cache is written to the config directory (not the workspace). Choose client_credentials (app-only) or device_code auth per your registration.

[notion] / [jira] — project integrations

[notion]
enabled = true
api_key = "secret_..."       # or NOTION_API_KEY

[jira]
enabled = true
base_url = "https://your-org.atlassian.net"
email = "[email protected]"
api_token = "..."            # or JIRA_API_TOKEN
allowed_actions = ["get_ticket", "search", "add_comment"]
timeout_secs = 30

jira.allowed_actions restricts the operations the tool may perform. Works against Atlassian Cloud or Server.

Note

Notion appears twice in the schema: [notion] configures the notion tool (query/read/write pages on demand), while [channels_config.notion] configures the Notion channel (polls a database as a task queue). They are independent.

[linkedin] — content publishing

Backs the linkedin tool. Credentials come from LINKEDIN_* env vars in .env.

[linkedin]
enabled = true
api_version = "202602"

[linkedin.content]
# automated-posting strategy settings

[linkedin.image]
# image-publishing settings

Media & vision sections

These configure speech, audio, image, and inbound-media handling. [transcription] and [tts] are shared by every channel that accepts or emits audio.

[transcription] — speech-to-text

[transcription]
enabled = true
default_provider = "groq"    # groq | openai | deepgram | assemblyai | google | local_whisper
api_key = "..."              # falls back to GROQ_API_KEY for Groq
model = "whisper-large-v3-turbo"
language = "en"              # optional ISO-639-1 hint
max_duration_secs = 120

Key	Default	Meaning
enabled	false	Enable transcription.
default_provider	"groq"	groq, openai, deepgram, assemblyai, google, local_whisper.
api_key	unset	Provider key; resolution falls back to URL-based env var → TRANSCRIPTION_API_KEY → GROQ_API_KEY → OPENAI_API_KEY.
model	"whisper-large-v3-turbo"	Groq default.
language	unset	ISO-639-1 hint.
initial_prompt	unset	Bias transcription toward expected vocabulary.
max_duration_secs	120	Skip longer audio.
transcribe_non_ptt_audio	false	Also transcribe non-voice-note audio on WhatsApp.

Per-provider sub-sections exist: [transcription.openai], [transcription.deepgram], [transcription.assemblyai], [transcription.google], [transcription.local_whisper] (url required, plus bearer_token, max_audio_bytes default 25 MB, timeout_secs default 300). Max audio size is 25 MB; supported formats include flac, mp3, mp4, m4a, ogg, opus, wav, webm.

[tts] — text-to-speech

[tts]
enabled = true
default_provider = "openai"  # openai | elevenlabs | google | edge | piper
default_voice = "alloy"
default_format = "mp3"       # mp3 | opus | wav
max_text_length = 4096

[tts.openai]
model = "tts-1"
speed = 1.0

[tts.elevenlabs]
model_id = "eleven_monolingual_v1"
stability = 0.5
similarity_boost = 0.5

Per-provider sub-sections: [tts.openai] (model, speed), [tts.elevenlabs] (model_id, stability, similarity_boost), [tts.google] (language_code), [tts.edge] (binary_path, free, subprocess-based), [tts.piper] (api_url, local). Provider keys fall back to OPENAI_API_KEY, ELEVENLABS_API_KEY, GOOGLE_TTS_API_KEY.

[media_pipeline] — automatic media understanding

Pre-processes inbound attachments before the agent sees them, prepending [Audio transcription: ...], [Image: <description>], or [Video summary: ...]. Uses [transcription] for audio and the active provider’s vision capability for images.

[media_pipeline]
enabled = true
transcribe_audio = true
describe_images = true
summarize_video = true

Falls back gracefully when the provider lacks vision ([Image: attached]). Defaults to enabled = false.

[multimodal] — image attachment handling

Controls how images are attached to agent messages.

[multimodal]
max_images = 4              # clamped 1–16
max_image_size_mb = 5       # clamped 1–20
allow_remote_fetch = false
vision_provider = "ollama"  # optional: route images to a dedicated vision model
vision_model = "llava:7b"

Supported image markers: [IMAGE:/path/to/file.png] or [IMAGE:data:image/png;base64,...]. Remote URLs are only fetched when allow_remote_fetch = true.

[image_gen] — image generation

Backs the image_gen tool (fal.ai Flux models by default; the standalone section also supports DALL-E, Stability, and Imagen providers).

[image_gen]
enabled = true
default_model = "fal-ai/flux/schnell"
api_key_env = "FAL_API_KEY"     # name of the env var holding the provider key

Key	Meaning
enabled	Enable the tool.
default_model	Default provider model ID.
api_key_env	Name of the environment variable that holds the provider API key.

Operational tool sections

These gate the analysis, pipeline, and workspace-maintenance tools.

[pipeline] — multi-step tool pipelines

Backs the execute_pipeline meta-tool.

[pipeline]
enabled = true
max_steps = 20
allowed_tools = ["file_read", "file_write", "shell"]

Steps run sequentially with {{step[N].result}} interpolation, or concurrently with parallel: true (no interpolation in parallel mode). allowed_tools restricts which tools a pipeline step may call.

[backup] — workspace backups

Backs the backup tool (create / list / verify / restore). Enabled by default.

[backup]
enabled = true
include_dirs = ["state", "sessions"]
max_keep = 10

Key	Default	Meaning
enabled	true	Enable backup operations.
include_dirs	—	Directories included in each backup.
max_keep	—	Retention count; older backups are pruned.

[data_retention] — purge & storage stats

Backs the data_management tool (stats / purge / list). Disabled by default; purge is destructive.

[data_retention]
enabled = true
retention_days = 90

[cloud_ops] — cloud advisory tools

Backs the read-only cloud_ops and cloud_patterns tools (IaC review, migration assessment, cost analysis, Well-Architected review). These never create or modify cloud resources.

[cloud_ops]
enabled = true

[security_ops] — managed cybersecurity

Backs the security_ops tool (alert triage, playbook execution, vulnerability parsing, reporting).

[security_ops]
enabled = true

Caution

Sensitive values across these sections — channel bot_tokens, integration API keys, the Nostr private_key — are encrypted at rest when secrets encryption is on (default during revka onboard). See Config: gateway, memory, security & platform for the [secrets] section.

Related pages

Configuration overview File location, precedence, and the schema export command.

Config: provider, agent & routing Provider/model keys, agent loop, pacing, and routing.

Config: gateway, memory, security & platform Gateway, tunnel, proxy, memory, autonomy, and secrets.

Channels overview How channels work plus per-platform setup runbooks.

Tools overview The full tool catalog and what each tool does.

Productivity integration tools Notion, Jira, Google Workspace, Microsoft 365, Composio, LinkedIn.