← Electerm Blog

AI Features in Electerm - Command Help, Smart Bookmarks, Themes and Agent Mode

2026-09-01
aillmagent-modebookmarksthemesproductivity

AI Features in Electerm

Electerm treats AI as a terminal coworker, not a chatbot bolted on the side. Configure any LLM API once, then use it for command help, bookmark generation, theme design, and autonomous multi-step tasks — all from the AI chat side panel, without leaving your sessions.

Base docs: AI Configuration Guide, Create Bookmark by AI, system prompt for any LLM chat, MCP Widget Usage Guide.

Security note: descriptions you send to a third-party AI API travel over the internet and may be logged by the provider. Never paste real passwords, private keys, or API secrets into AI prompts unless you run a local/trusted model.

Related videos:

1. Connect any model in 2 minutes

Electerm speaks three LLM wire formats, so almost every provider works:

  • OpenAI Chat Completions (/chat/completions) — default
  • OpenAI Responses (/responses)
  • Anthropic Messages (/messages)

Built-in presets fill URL/model for OpenAI, DeepSeek, OpenRouter, Gemini (OpenAI-compat endpoint), Groq, Together, Mistral, xAI/Grok, Perplexity, Moonshot/Kimi, SiliconFlow, AtlasCloud and more. You only add the API key. Advanced knobs: custom auth header (Authorization: Bearer / x-api-key / ...), system role, response language, proxy (socks5://...), config history (last 20), test-connection button, local suggestion cache.

Free option: ai.electerm.org gives electerm users a free API (https://ai.electerm.org/api/ai + /chat/completions). Sign in with GitHub → copy key → paste into electerm AI settings → Save → open the side panel. It is run by the electerm developer on a free-tier account, so heavy use may hit limits — then plug your own key from Mistral/OpenRouter/etc. (free-llm list).

Full field reference: AI Configuration Guide.

2. AI-assisted commands

The daily driver:

  • Generate: describe the goal in plain language ("find the 10 largest docker images and prune dangling ones, explain each flag") → paste-ready command with a short explanation.
  • Explain selection: highlight cryptic output or a one-liner from Stack Overflow → AI explains what it does before you run it.
  • Fix errors: paste the failing command + stderr → corrected version.
  • Script writing: multi-line bash/python snippets with markdown formatting, honoring your configured system role ("terminal expert, brief explanations, markdown").

Suggestions are cached locally to save API calls. Set response language once and every answer follows it.

3. AI-driven bookmark creation

Bookmarks have ~40 fields (tunnels, hops, X11, encodings, run-scripts...). Instead of filling forms, describe the machine:

SSH to 192.168.0.10 as admin, enable X11, forward remote port 5900
to local 5900, start in /home/admin, use the ssh-agent, and add a
tunnel 127.0.0.1:8080 → remote 10.0.0.5:80.

Click Create Bookmark by AI → Generate → review the pretty-printed JSON (edit / copy / download) → Confirm. Arrays create multiple bookmarks in batch. Prefer ChatGPT/Claude's website? Use the portable system prompt or the hosted bookmark generator — same schema, no install.

Video: Electerm Use AI to Create Bookmarks. Guide: Create Bookmark by AI.

4. AI theme customization

Describe a vibe — "solarized dark with amber accents, high-contrast cursor, 14px JetBrains Mono" — and let the model draft the theme JSON, then fine-tune in the visual theme editor. Import community themes from the iTerm2 collection or publish yours at theme.electerm.org (live preview + AI creation built in).

Video: Electerm Theme Settings and Editing.

5. Agent mode: AI that runs tasks in the terminal

Chat answers text; agent mode acts. Give a goal ("upgrade node 18→20 on staging-web-02, run tests, roll back on failure"), and the agent plans shell steps, runs them in the terminal, reads output, and continues — with your approval gates. Combined with electerm's batch execution, one goal can fan out across many hosts. Start with read-only recon tasks, then graduate to mutating runs once you trust the loop.

6. AI assistant + MCP widgets

The side-panel AI assistant stays docked next to your sessions: ask follow-ups while a long command runs, summarize logs, draft commit messages from git diff. Power users can extend it via the MCP (Model Context Protocol) widget — expose external tools and context sources to the assistant. Setup: MCP Widget Usage Guide.

Practical setup recipe

  1. AI settings → Preset = your provider → paste key → Test Connection → Save.
  2. System role: Terminal expert, provide commands for different OS, explain usage briefly, use markdown format.
  3. Language = your language. Proxy only if your network needs it.
  4. Pin the AI panel; try: generate → explain-selection → bookmark-from-text → theme draft → agent recon task.
  5. Clear the suggestion cache if answers go stale.

Where next