> ## Documentation Index
> Fetch the complete documentation index at: https://comis-feature-matrix-channel.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.
# Platform Tools
> Overview of all Comis-specific tools agents can use for messaging, scheduling, and administration
Platform tools are built specifically for Comis. They let agents do things
like schedule tasks, send messages across channels, manage sessions, analyze
media, control infrastructure, and perform platform-specific actions. Unlike
built-in tools (which handle files and web access), platform tools are about
working within the Comis ecosystem.
Platform tools make up the bulk of Comis's **70 total built-in tools**. The
[built-in tools page](/skills/built-in-tools) covers the framework-level
capabilities (file ops, shell, web, browser); everything below is what Comis
adds on top.
## Tool Categories
Every platform tool is documented in detail in the [Agent Tools](/agent-tools/index)
tab. Here is a summary of what is available, organized by category:
| Category | Tools | What They Do | Details |
| ---------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------- | ------------------------------------------------------- |
| Scheduling | `cron`, `heartbeat_manage` | Create scheduled jobs, manage heartbeat timing | [Scheduling Guide](/agent-tools/scheduling) |
| Messaging | `message`, `notify_user` | Send, reply, react, edit, delete messages; proactive notifications | [Messaging Guide](/agent-tools/messaging) |
| Sessions | `session_search`, `session_status`, `sessions_list`, `sessions_history`, `sessions_send`, `sessions_spawn`, `subagents` | Manage conversations and spawn sub-agents | [Sessions Guide](/agent-tools/sessions) |
| Orchestration | `pipeline` | Define, execute, and manage multi-agent DAG workflows | [Execution Graphs](/agents/execution-graphs) |
| Media | `image_analyze`, `tts_synthesize`, `transcribe_audio`, `describe_video`, `extract_document`, `image_generate` | Analyze images, generate speech, transcribe audio, process documents, generate images | [Media Guide](/agent-tools/media) |
| Infrastructure | `gateway`, `providers_manage` | Read and modify configuration, manage the gateway and LLM providers | [Infrastructure Guide](/agent-tools/infrastructure) |
| Memory | `memory_get`, `memory_manage`, `memory_search`, `memory_store` | Search, read, store, and manage long-term memories | [Sessions Guide](/agent-tools/sessions) |
| Context | `session_search` | Search the current session's conversation history | [Context Tools](/skills/built-in-tools#context-tools) |
| Browser | `browser` | Automate a headless browser with 16 actions | [Browser Guide](/agent-tools/browser) |
| Background | `background_tasks` | List, inspect, cancel, and read output from background tasks | Below |
| Fleet Management | `agents_manage`, `obs_query`, `sessions_manage`, `memory_manage`, `channels_manage`, `tokens_manage`, `models_manage`, `providers_manage`, `skills_manage`, `mcp_manage`, `heartbeat_manage` | Supervisor-level administration tools | [Infrastructure Guide](/agent-tools/infrastructure) |
| Platform Actions | `discord_action`, `telegram_action`, `slack_action`, `whatsapp_action` | Platform-specific moderation (54 total actions) | [Platform Actions Guide](/agent-tools/platform-actions) |
### Scheduling
The `cron` tool lets agents create, list, update, and remove scheduled jobs.
Agents can set up recurring tasks using cron expressions (like "every day at
9am"), interval-based schedules ("every 30 minutes"), or one-time future runs.
This is how agents automate routine work -- daily reports, periodic check-ins,
scheduled reminders, and more.
The `heartbeat_manage` tool controls the agent's periodic heartbeat -- an
automatic wake-up that lets agents check for pending work even when no one has
messaged them. Agents can adjust the heartbeat interval or disable it entirely.
### Messaging
The `message` tool is how agents communicate across channels. It supports
sending new messages, replying to specific messages, adding reactions, editing
previous messages, and deleting messages. A single agent can message across
different platforms -- posting to Discord and Telegram from the same
conversation.
This is one of the most frequently used platform tools. Whenever an agent
needs to proactively reach out (not just reply), it uses the `message` tool.
### Sessions
Session tools give agents awareness of their conversation state and the
ability to collaborate with other agents:
* `session_search` -- search session history by query
* `session_status` -- get the current session's status
* `sessions_list` -- list all sessions for this agent
* `sessions_history` -- retrieve session history entries
* `sessions_send` -- send a message to another agent's session
* `sessions_spawn` -- create a new sub-agent session for parallel work
* `subagents` -- manage spawned sub-agents
Fleet listing is available via `agents_manage` (action: `list`).
### Orchestration
The `pipeline` tool lets agents define execution graphs (DAGs) of sub-agent
tasks with dependency ordering, barrier modes, budget controls, and shared
data folders. Agents can define, execute, monitor, cancel, retrieve outputs,
save, load, list, delete, and synthesize-from-intent graphs -- 10 actions in
total.
See [Execution Graphs](/agents/execution-graphs) for concepts and the
[Pipeline Tool Guide](/agent-tools/pipelines) for the full action reference.
### Media
Media tools handle rich content beyond text:
* `image_analyze` -- describe images or answer questions about visual content
* `tts_synthesize` -- generate speech audio from text (text-to-speech)
* `transcribe_audio` -- convert speech to text (speech-to-text)
* `describe_video` -- summarize video content
* `extract_document` -- pull text from PDFs and other document formats
These tools integrate with external AI services (such as OpenAI, ElevenLabs,
Deepgram, and Groq) depending on your configuration.
### Infrastructure
The `gateway` tool lets agents read and modify the Comis configuration at
runtime. Agents with appropriate permissions can inspect current settings,
apply configuration patches, and manage gateway state. This is particularly
useful for supervisor agents that need to reconfigure the system -- adjusting
model settings, changing rate limits, or updating channel connections --
without human intervention.
### Memory
Memory tools give agents long-term recall beyond the current conversation:
* `memory_store` -- save important information with metadata tags so it can
be found later
* `memory_search` -- retrieve relevant memories using semantic search, which
finds memories by meaning rather than exact keywords
* `memory_get` -- retrieve a specific memory by its identifier
Memories persist across sessions, so an agent that stores a user's preferences
today can recall them weeks later in a completely different conversation.
### Context
In the default **DAG** (LCD) mode, three **in-session** expansion tools
(`ctx_search`, `ctx_inspect`, `ctx_expand`) recover detail the summary hierarchy
compressed away within the current conversation -- available in DAG mode
(`contextEngine.version` defaults to `"dag"`), `never-export`, and distinct from
cross-session recall (`memory_search`, `session_search`). In the opt-in
**pipeline** mode, `session_search` searches the raw session history instead
(case-insensitive substring match, role-scoped). See
[Context expansion tools](/agent-tools/sessions#context-expansion-tools)
and [Context Tools](/skills/built-in-tools#context-tools) for the `session_search`
parameter reference.
### Fleet Management
Fleet management tools are supervisor-level capabilities for managing the
entire Comis installation:
* `agents_manage` -- six actions for agent lifecycle: `create`, `get`, `update`, `delete`, `suspend`, `resume` (create/delete take effect immediately without daemon restart; update requires restart)
* `obs_query` -- query operational data with the action sub-system: `diagnostics`, `billing`, `delivery`, and `channels` (each action also accepts a `sub_action` for finer queries)
* `sessions_manage` -- administrative session operations
* `memory_manage` -- administrative memory operations (bulk delete, export)
* `channels_manage` -- manage channel connections and settings
* `tokens_manage` -- manage authentication tokens
* `models_manage` -- manage model configurations and API keys
* `providers_manage` -- manage LLM provider configurations (list, get, create, update, delete, enable, disable). Admin trust required. Create and delete require approval. Providers include cloud endpoints (NVIDIA NIM, Groq, DeepSeek), local inference (Ollama, vLLM, LM Studio), and multi-provider gateways (OpenRouter).
* `skills_manage` -- manage skill registrations (supports per-agent scope: local or shared)
* `mcp_manage` -- manage MCP server connections
* `heartbeat_manage` -- adjust agent heartbeat configuration at runtime
These tools are restricted to supervisor agents. See the section below on
regular vs supervisor tools for details.
### Platform Actions
Each supported messaging platform has its own action tool:
* `discord_action` -- Discord-specific moderation and administration
* `telegram_action` -- Telegram group and channel management
* `slack_action` -- Slack workspace and channel operations
* `whatsapp_action` -- WhatsApp messaging and group management
These provide platform-specific capabilities like banning users, managing
channels, pinning messages, setting permissions, and other actions unique to
each platform. There are 54 total actions across all four platforms.
### Unified Tools
Comis provides unified versions of several tool families that consolidate
multiple individual tools into a single tool with an `action` parameter.
These reduce the number of tools in the agent's context while keeping the
same functionality.
#### Session tools
Four individual tools for querying and inspecting sessions:
* **`session_search`** -- search session history by query (keywords, phrases, or boolean expressions); supports role-scoped filtering and optional LLM summarization.
* **`session_status`** -- get the current session's status, including model, context usage, and active configuration.
* **`sessions_history`** -- retrieve history entries for a target session key, with pagination (`offset`, `limit`).
* **`sessions_list`** -- list all sessions for this agent, filterable by kind (`dm`, `group`, `sub-agent`) and recency (`since_minutes`).
#### Memory tools
Four individual tools for long-term memory operations:
* **`memory_get`** -- retrieve a memory entry by key (file path relative to workspace), with optional line range.
* **`memory_manage`** -- administrative memory operations: `stats`, `browse`, `delete`, `flush`, and `export`. Supports tenant and agent ID scoping.
* **`memory_search`** -- semantic search over memory using a natural-language query; returns ranked results by meaning rather than exact keywords.
* **`memory_store`** -- store a new memory entry with optional tags for later retrieval.
### Notifications
#### notify\_user
Sends a proactive notification to the user outside of the normal
request-response flow. Supports priority levels and automatic channel
resolution. The daemon-side handler applies rate limiting, deduplication,
quiet hours enforcement, and channel resolution guards before delivery.
**Parameters:**
| Parameter | Type | Required | Description |
| -------------- | -------- | -------- | -------------------------------------------------------------------------------------------------- |
| `message` | `string` | Yes | Notification text to send to the user |
| `priority` | `string` | No | Priority level: `low`, `normal` (default), `high`, or `critical`. `critical` bypasses quiet hours. |
| `channel_type` | `string` | No | Target channel type (e.g., `telegram`, `discord`). Omit for auto-resolution. |
| `channel_id` | `string` | No | Target channel/chat ID. Required when `channel_type` is specified. |
### Image Generation
#### image\_generate
Generates images from text prompts using configured image generation
providers (FAL or OpenAI). Generated images are delivered directly to the
current channel. The daemon-side handler applies rate limiting and safety
checking before executing the generation request.
**Parameters:**
| Parameter | Type | Required | Description |
| --------- | -------- | -------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `prompt` | `string` | Yes | Text description of the image to generate |
| `size` | `string` | No | Image size. Provider-specific: fal.ai uses presets (`square_hd`, `landscape_16_9`), OpenAI uses pixel dimensions (`1024x1024`, `1792x1024`). Omit for provider default. |
### Background Tasks
#### background\_tasks
Manages background tasks that are running or have completed for the current
agent. Provides four actions for task lifecycle management. Any agent can
check its own tasks -- this is not restricted to supervisor agents.
**Parameters:**
| Parameter | Type | Required | Description |
| --------- | -------- | --------------------------- | ----------------------------------------- |
| `action` | `string` | Yes | `list`, `get`, `cancel`, or `read_output` |
| `taskId` | `string` | For get/cancel/read\_output | Task ID to operate on |
**Actions:**
* **list** -- show all tasks for the current agent with status and metadata
* **get** -- get detailed information about a specific task by ID
* **cancel** -- cancel a running task
* **read\_output** -- read the output of a completed task
## Regular vs Supervisor Tools
Most platform tools are available to any agent running with the `coding`,
`messaging`, or `full` tool policy profile. However, fleet management tools
require the `supervisor` tool policy profile or an explicit `allow` list in
the tool policy configuration. The supervisor-restricted tools are:
* `agents_manage`
* `obs_query`
* `sessions_manage`
* `memory_manage`
* `channels_manage`
* `tokens_manage`
* `models_manage`
* `providers_manage`
* `skills_manage`
* `mcp_manage`
* `heartbeat_manage`
These are admin-level tools for managing the entire Comis installation -- you
would not want every agent to have access to them. A typical setup has one
supervisor agent with full access and several regular agents with only the
tools they need for their specific role.
See [Tool Policy](/skills/tool-policy) for details on configuring which tools
each agent can use.
## Web and Browser Tools
The `web_search`, `web_fetch`, and `browser` tools are technically built-in
tools (they come from the underlying coding agent framework), not platform
tools. However, they are documented in the Agent Tools tab because most users
expect to find web and browser capabilities alongside other tool guides. If
you are looking for details on web search providers or browser automation
actions, you will find them there:
* [Web Tools Guide](/agent-tools/web-tools) -- 8 web search providers and URL
content fetching
* [Browser Guide](/agent-tools/browser) -- headless browser with 16 automation
actions
## Next Steps
See the master reference table of all tools
Control which tools agents can use
File, shell, and web tools
Back to the skills taxonomy