Comis is a self-hosted, open-source AI assistant platform that lives inside your messaging apps -- not a browser tab. Deploy a team of specialized agents with persistent memory, 50+ tools, and DAG-based workflows across 9 platforms. Secured by 22 defense layers and optimized to reduce LLM costs by 80%+.

comis (Latin) -- courteous, kind, affable, gracious. That's what an AI assistant should be.

Why Comis?

Security: the LLM is the attack surface

AI agents have access to your messaging apps, files, shell, and API keys. A single prompt injection can make the LLM leak secrets, execute destructive commands, or exfiltrate private data.

Comis assumes the LLM will be attacked. 22 independent defense layers intercept threats at every stage: input scanning for injection patterns, output scanning for leaked secrets, kernel-enforced exec sandboxes (Bubblewrap on Linux, sandbox-exec on macOS), per-agent tool restrictions with approval gates, trust-partitioned memory, canary tokens for prompt extraction detection, AES-256-GCM encrypted secrets, SSRF guards, and a 14-point security audit CLI.

Cost: prompt caching saves 81%

Comis has the most advanced prompt cache management available for Anthropic, with 20 dedicated optimizations including adaptive TTL escalation, cache fence protection, and sub-agent spawn staggering. Gemini gets native CachedContent API integration. OpenAI is supported with completion storage.

| Metric | Value | |---|---| | 76-call Opus 4.6 session | $5.02 vs $26.42 uncached | | Cache hit rate | 94% of input tokens | | 8-agent trading pipeline | $2.11 for 788K tokens |

Context: scales without degradation

Most assistants silently drop old messages when context fills up. Comis never deletes a message. An 8-layer context engine handles compaction, rehydration, dead content eviction, and progressive tool disclosure -- keeping 50+ tools available without burning context. Between sessions, a background learning job extracts preferences and facts from past conversations. The agent gets better over time.

Orchestration: team agents from natural language

"Have four analysts research NVDA in parallel, then run a bull vs bear debate, and let the head trader make the final call."

One sentence creates a 7-node DAG pipeline with parallel fan-out, multi-round debate, and synthesis. No YAML, no scripting. 7 node types (agent, debate, vote, refine, collaborate, map-reduce, approval gate), 3-tier concurrency control, and barrier synchronization.

Quick Start

One-liner -- installs Node.js and everything else:

curl -fsSL https://comis.ai/install.sh | bash

Or with npm (requires Node.js 22+):

npm install -g comisai
comis init               # interactive setup wizard
comis daemon start       # start the daemon

The wizard walks you through provider selection, API key validation, channel configuration, and daemon startup in under 5 minutes. Message your agent. That's it.

Features

| | Feature | Description | |---|---|---| | | 9 messaging channels | Telegram, Discord, Slack, WhatsApp, Signal, iMessage, IRC, LINE, Email -- text, voice, images, files, reactions, threads | | | Multi-agent fleet | Specialized agents with isolated memory, budgets, and tool policies. Per-agent model selection. Sub-agent spawning. | | | Persistent memory | SQLite + FTS5 + vector search with trust-partitioned storage. RAG retrieval keeps context relevant weeks later. | | | DAG pipelines | 7 node types built from natural language. Parallel fan-out, debate, vote, refine, map-reduce, approval gates. | | | MCP ecosystem | 50+ tool servers -- GitHub, Gmail, Notion, databases, browser automation, shell. Add any MCP server with one config line. | | | Skills system | Modular prompt packages for domain expertise, workflows, and persona traits. Runtime eligibility filtering, live reload. | | | 8-layer context engine | Dead content eviction, LLM compaction, rehydration, progressive tool disclosure -- 80%+ cost reduction. | | | 22 security layers | Exec sandbox, injection detection (40+ patterns), AES-256 encrypted secrets, SSRF guard, canary tokens, approval gates. | | | Any model, any provider | Claude, GPT, Gemini, Groq, Ollama, OpenRouter, and more. Tool presentation adapts to each model's context window. | | | Media processing | Vision, STT (Whisper/Groq/Deepgram), TTS (OpenAI/ElevenLabs/Edge), image generation (FAL/Gemini/DALL-E), PDF extraction. | | | Headless browser | Web automation, screenshots, form filling, JavaScript execution via Playwright. | | | Scheduling & cron | Recurring tasks, heartbeat health checks, cron triggers for pipeline automation. | | | Gateway & API | OpenAI-compatible API, JSON-RPC, WebSocket, mTLS, bearer auth. | | | Observability | Structured Pino logging, circuit breakers, per-agent cost tracking, trace IDs. |

Supported Channels

| Platform | Text | Media | Voice | Threads | Reactions | Groups | |----------|:----:|:-----:|:-----:|:-------:|:---------:|:------:| | Telegram | x | x | x | x | x | x | | Discord | x | x | x | x | x | x | | Slack | x | x | - | x | x | x | | WhatsApp | x | x | - | - | x | x | | Signal | x | x | - | - | - | x | | LINE | x | x | - | - | - | x | | iMessage | x | x | - | - | - | x | | IRC | x | - | - | - | - | x | | Email | x | x | - | x | - | - |

CLI

comis init               # Interactive setup wizard
comis configure          # Update configuration
comis daemon start|stop  # Manage the background daemon
comis pm2 setup|start    # PM2 process management
comis status             # Show current state
comis health             # Verify setup
comis doctor             # Diagnose and repair issues
comis security audit     # Run 14-point security audit
comis secrets            # Manage encrypted credentials
comis channel            # Add or remove messaging platforms
comis models             # List and switch LLM providers
comis agent              # Manage agent identity and behavior
comis memory             # Search and manage persistent memory
comis sessions           # List and manage conversations

Architecture

Hexagonal (ports and adapters). Core defines port interfaces -- adapters implement them. Swap Discord for Matrix, SQLite for Postgres, or OpenAI for Ollama without touching core logic. 19 ports, 30+ adapters, 13 packages. Every function returns Result<T, E> -- zero thrown exceptions.

shared (Result type, utilities)
 └── core (domain types, ports, event bus, security, config)
      ├── memory (SQLite + FTS5 + vector search)
      ├── gateway (HTTP, JSON-RPC, WebSocket, mTLS)
      ├── skills (tools, MCP, media processing)
      ├── scheduler (cron, heartbeats)
      ├── agent (executor, budget, RAG, sessions)
      ├── channels (9 platform adapters)
      ├── cli (commands, RPC client)
      └── daemon (orchestrator, observability)

Requirements

• Node.js >= 22
• Linux (primary target; macOS for development)
• Native build tools for SQLite and Sharp

Links

License

Apache-2.0