Master reference for every tool your Comis agents can use
Comis ships with 73 built-in agent tools spanning file operations, web access, messaging, scheduling, media processing, browser automation, context management, interactive terminal sessions, and administration. This page is the master reference — every tool listed with its category, description, and link to detailed documentation.When an agent receives a message, it decides which tools to use based on what you asked it to do. Need a daily briefing? The agent uses the cron tool. Want to send a message to another channel? It uses message. Need to analyze an image someone sent? It reaches for image_analyze.The tools listed below are everything your agents have at their disposal. Each tool is linked to its detailed guide page where you will find parameters, examples, and tips for effective use.
Core capabilities shipped with Comis — file operations, shell commands, web search, browser automation, messaging, scheduling, sessions, media, infrastructure, context, interactive terminal sessions, supervisor admin, and platform actions. Available on every agent unless restricted by tool policy.
MCP tools
Configurable
External Model Context Protocol servers — Anthropic’s reference servers, community-built MCP servers, and your own. Discovered at runtime and qualified with mcp:{server}/{tool} names. See MCP Management.
Custom skills
Unlimited
Markdown-defined “mini-programs” with optional tool restrictions and bin/env requirements. Comis ships 7 built-in skills; you can author your own. See Skills.
Tool access is controlled by tool policy. By default, agents have access to all tools, but you can restrict access per agent using profiles (like coding or messaging) or fine-grained allow/deny lists.
All tool names use snake_case — for example, web_search, sessions_spawn, and memory_store. When referencing tools in configuration files (such as tool policy allow/deny lists), use these exact names.
File and code tools are always available unless explicitly denied by tool policy. They can be individually toggled in config.yaml under the builtinTools section.
Tool
Description
Guide
read
Read file contents with line numbers and pagination
Interactive terminal sessions for driving allowlisted binaries (added in v2.11). Requires an operator-configured allowEntries allowlist — without a matching entry the tool rejects. Requires bubblewrap sandbox on Linux. See exec sandbox.
Tool
Description
Guide
terminal_session_create
Start an interactive terminal session driving an allowlisted binary.
Scheduling tools let agents create automated tasks that run on a timer. Agents can set up recurring cron jobs, one-time reminders, and manage their own heartbeat — a periodic wake-up that triggers routine maintenance or check-in tasks.
The message tool works across all connected channels — Discord, Telegram, Slack, WhatsApp, and every other supported platform. Agents can send new messages, reply to specific messages, add emoji reactions, edit or delete messages, fetch message history, and send file attachments.Agents can also send messages to channels they are not currently conversing in, making the message tool ideal for notifications, alerts, and cross-channel coordination.
Tool
Description
Guide
message
Send, reply, react, edit, delete, fetch, and attach messages
Session tools let agents manage their conversations, communicate with other agents, and spawn sub-agents for specialized tasks. For example, a lead agent can spawn a researcher sub-agent and a writer sub-agent to work in parallel.
The pipeline tool defines and runs multi-agent execution graphs (DAGs). Nodes run in parallel when independent, or wait for upstream nodes via depends_on. Comis ships 7 built-in node type drivers — agent, debate, vote, refine, collaborate, approval-gate, and map-reduce.
Tool
Description
Guide
pipeline
Define, execute, save, load, list, delete, monitor, cancel, and read outputs from execution graphs (9 actions)
Media tools process images, audio, video, and documents. Agents can analyze images with vision AI, convert text to speech, transcribe voice messages, describe video content, and extract text from documents like PDFs. These tools require appropriate AI providers to be configured (for example, vision capabilities for image analysis, and STT providers for transcription).
Tool
Description
Guide
image_analyze
Vision AI image analysis (file, URL, base64, attachment)
Memory tools give agents access to their long-term memory. Agents automatically build up memory over time, but these tools let them explicitly search, read, and store information. Memory uses hybrid text and vector search for semantic matching, so agents can find relevant memories even with different wording.
Grounded natural-language answer over the agent’s recall pipeline. Default-on; disable per-agent with dialectic.enabled: false, or via the memory.costFeatures.enabled: false kill switch.
In-session recovery tools over the DAG (LCD) context engine’s lossless store.
Available when the agent runs in DAG mode (contextEngine.version: "dag").
They recover detail the summary hierarchy compressed away within this
conversation — they are distinct from cross-session recall
(memory_search / session_search), which they share no code with, and they
never reach other sessions or tenants. All three are never-export (never
reachable via the MCP server) and read-only. See
Context Management.
Tool
Description
Documentation
ctx_search
Full-text search over THIS conversation’s compressed history (FTS with a LIKE fallback).
These platform-management tools let agents inspect or modify the running Comis daemon. The gateway tool drives configuration and the secret store; background_tasks inspects the long-running task queue; mcp_manage reaches into the MCP client manager. See also: browser (listed under the Web and Browser section).
Tool
Description
Guide
gateway
Configuration management and gateway control (11 actions: read, patch, apply, restart, schema, status, history, diff, rollback, env_set, env_list)
Supervisor tools require admin-level trust. They manage the entire platform, not just the current agent’s session. Only agents with the supervisor profile or explicit allow entries can use these tools.
Supervisor tools can modify agents, delete sessions, flush memory, and manage gateway tokens. Grant these tools only to trusted agents that need fleet-wide control. See tool policy for how to restrict access.
Platform action tools perform moderation and administration tasks specific to each messaging platform. Each tool wraps the native API of its platform, letting agents kick users, manage roles, pin messages, create channels, and more — all from within a conversation.The available actions vary by platform. See the Platform Actions guide for the full list of actions per platform.
The message tool works across all 9 connected channels, but not every action is supported everywhere. The matrix below summarizes per-channel capabilities — consult the Channels Overview for full per-channel detail.
Channel
Send
Reply
React
Edit
Delete
Fetch History
Threads
Streaming
Discord
Yes
Yes
Yes
Yes
Yes
Yes
Native
Edit
Telegram
Yes
Yes
Yes
Yes
Yes
No
Forum topics
Edit
Slack
Yes
Yes
Yes
Yes
Yes
Yes
Reply chain
Edit
WhatsApp
Yes
Yes (quoted)
Yes
Yes
Own only
No
None
Block
Signal
Yes
Yes (quoted)
Yes
No
Yes
No
None
Block
iMessage
Yes
No
No
No
No
Yes
None
None
LINE
Yes
Yes (push)
No
No
No
No
None
None
IRC
Yes
Mention only
No
No
No
No
None
None
Email
Yes
Yes (headers)
No
No
No
No
Reply chain
None
“Streaming” indicates how live agent output is delivered: Edit means the agent edits a placeholder message as content streams in; Block means the agent buffers and emits complete chunks; None means no streaming support (the agent waits for the full reply).
Not sure which tool you need? Start here. The table below maps common tasks to the specific tool and action that handles them. Each row links to the detailed guide page where you can find parameters, examples, and tips.
