Yes, for the CLI. discli is installed via pip (it is a Python package), but once installed, the discli binary works as a standalone command-line tool. You do not need to write any Python.

For agents, any language works. The discli serve protocol is language-agnostic JSONL over stdin/stdout. Your agent can be written in Node.js, Go, Rust, Ruby, Bash, or anything else that can spawn a subprocess and read/write lines.

# Install (requires Python/pip one time)
pip install discord-cli-agent

# Use from any language — no Python needed after install
discli message send "#general" "Hello from Rust!"

No. discli uses the official Discord API exclusively. Self-hosted Discord alternatives (Fosscord, Spacebar, Revolt, Guilded) have different APIs and are not supported.

If a self-hosted platform offers a Discord-compatible API, discli might work by pointing the API base URL at it, but this is untested and unsupported.

One discli serve instance per token. Discord’s gateway allows a single active WebSocket connection per bot token. Running two discli serve processes with the same token causes a reconnect loop.

However, CLI commands can run concurrently with no issues. They use stateless REST API calls, so you can have one discli serve running and simultaneously use discli message send from another terminal or script.

If you need multiple independent agents, create a separate bot application and token for each one.

No. discli uses the official Discord Bot API with proper bot tokens. It does not:

• Use user/selfbot tokens
• Scrape the Discord client
• Reverse-engineer proprietary protocols
• Bypass rate limits

discli is no different from any other Discord bot library (discord.py, discord.js) in terms of API compliance. Your bot must still follow Discord’s Developer Terms of Service and Developer Policy.

Yes, discli is model-agnostic. The JSONL protocol does not know or care what AI model (if any) is processing events. You can use:

• OpenAI (GPT-4, GPT-4o, etc.)
• Anthropic Claude (Claude 3, Claude 4, etc.)
• Local models (Llama, Mistral, Phi via Ollama, llama.cpp, vLLM)
• Any other model that can read JSON and produce text
• No model at all — discli works just fine for rule-based bots or simple scripts

# Your agent decides what to do with events
event = json.loads(line_from_discli)

# Send to whatever model you want
response = openai.chat(...)       # or
response = anthropic.messages(...) # or
response = ollama.generate(...)    # or
response = my_rules_engine(event)  # no AI needed

Text-only for now. discli currently supports sending and receiving text messages through the JSONL serve protocol. Rich embeds, buttons, select menus, and modals are not yet supported in serve mode.

For CLI usage, basic embeds are supported:

discli message send "#announcements" "New release!" \
  --embed-title "v2.1.0" \
  --embed-desc "Bug fixes and performance improvements"

Full component support (buttons, select menus, modals) in both CLI and serve mode is planned for a future release.

It depends on what you are doing. Here is a minimum set for common use cases:

Use pip’s upgrade flag:

pip install --upgrade discord-cli-agent

Check your current version:

discli --version

discli follows semantic versioning. Minor and patch updates are backward-compatible. Major version bumps may include breaking changes and will be documented in the changelog.

Yes. discli is a standard Python package with no system-level dependencies beyond Python itself. A minimal Dockerfile:

FROM python:3.11-slim
RUN pip install discord-cli-agent
ENV DISCORD_BOT_TOKEN=your_token_here
CMD ["discli", "serve", "--events", "messages"]

For production, use a non-root user and pass the token via environment variable at runtime (not baked into the image):

FROM python:3.11-slim
RUN pip install discord-cli-agent && \
    useradd -m discli
USER discli
ENTRYPOINT ["discli"]

docker run -e DISCORD_BOT_TOKEN=your_token my-discli-bot serve --events messages

Yes, two layers:

1. discli’s built-in rate limiter — 5 destructive actions (send, delete, ban, etc.) per 5-second window. This protects against runaway agents and keeps you well under Discord’s limits. Read-only actions (list, search, info) are not rate-limited by discli.

2. Discord’s API rate limits — vary by endpoint. Most endpoints allow 5 requests per 5 seconds per route. Some endpoints (like bulk message delete) have stricter limits. discli handles Discord 429 responses automatically with exponential backoff.

If you need higher throughput for a specific use case, check the Discord Rate Limits documentation for per-endpoint details.