Reads, edits, and runs code in any project. Use Ollama locally or connect to Groq, OpenRouter, Zen, and more.
chmod +x manthra-linux-x64 && mv manthra-linux-x64 ~/.local/bin/manthra
chmod +x manthra-linux-arm64 && mv manthra-linux-arm64 ~/.local/bin/manthra
chmod +x manthra-macos-x64 && mv manthra-macos-x64 ~/.local/bin/manthra
⚠ macOS Gatekeeper will block this binary
After downloading, run these commands in Terminal to allow it:
xattr -c ~/Downloads/manthra-macos-x64 codesign --force --deep --sign - ~/Downloads/manthra-macos-x64 chmod +x ~/Downloads/manthra-macos-x64 mv ~/Downloads/manthra-macos-x64 ~/.local/bin/manthra
Tip: the curl installer above handles this automatically.
Run the signing steps below before first launch.
⚠ macOS Gatekeeper will block this binary
After downloading, run these commands in Terminal to allow it:
xattr -c ~/Downloads/manthra-macos-arm64 codesign --force --deep --sign - ~/Downloads/manthra-macos-arm64 chmod +x ~/Downloads/manthra-macos-arm64 mv ~/Downloads/manthra-macos-arm64 ~/.local/bin/manthra
Tip: the curl installer above handles this automatically.
Place in any folder on your PATH, e.g. C:\Users\you\bin\manthra.exe
Or install via Git Bash or WSL2: curl -fsSL https://manthra.informaticsint.au/install | bash
macOS 12+ · Apple Silicon (M1 / M2 / M3 / M4) · ~150 MB
macOS 12+ · Intel · ~160 MB
Windows 10/11 · x64 · NSIS installer · ~150 MB
chmod +x Manthra-linux-x64.AppImage && ./Manthra-linux-x64.AppImage
chmod +x Manthra-linux-arm64.AppImage && ./Manthra-linux-arm64.AppImage
The desktop app includes the full Manthra interface — chat, multi-agent, file picker, and settings — no terminal needed.
Documentation
Run Manthra completely offline on your own machine. Free, private, no rate limits.
Download from ollama.ai or install with one command on macOS and Linux.
Choose a model from the Ollama library. For coding tasks, qwen2.5-coder and deepseek-r1 are strong choices.
Ollama listens on http://127.0.0.1:11434 by default. On macOS it starts automatically after installation.
The installer detects your OS and architecture, downloads the right binary, and adds it to your PATH.
Open the built-in web UI to add your Ollama instance, select a model, and save the configuration.
Navigate to any project directory and launch Manthra. It reads your files relative to where you started it.
/init inside Manthra to generate an AGENTS.md file for your project. Manthra loads it automatically and uses it as project context — architecture, conventions, how to build, etc.
Cloud setup
Run Ollama on a cloud VM or dedicated machine and connect Manthra from any device — no local GPU required.
Any cloud provider works — AWS, GCP, Azure, DigitalOcean, Hetzner, etc. A GPU instance gives best performance; CPU-only works for smaller models.
Set up a systemd service so Ollama starts automatically on reboot and stays running in the background.
Allow inbound TCP on port 11434. Then restrict it to only your IP if possible — never leave it open to the internet without authentication.
Open the web UI and add the remote server's IP or hostname as the base URL. The port is 11434 by default.
Put Nginx or Caddy in front of Ollama to get HTTPS and basic authentication. This lets you safely connect over the public internet.
Set OLLAMA_HOST before running Manthra to override the base URL without opening the web UI.
MCP
Connect any Model Context Protocol server to give Manthra new tools — browser automation, database access, GitHub/GitLab integration, and more.
MCP (Model Context Protocol) is an open standard for connecting AI assistants to external tools and data sources. Any MCP-compatible server can be registered with Manthra — the AI will see its tools and call them automatically when needed.
mcp__<server-name>__<tool-name> to avoid conflicts — e.g. mcp__playwright__browser_navigate.
Open the web config panel, navigate to MCP Servers, and click Add MCP Server. Fill in the name, transport (stdio for local processes, http for remote servers), and command or URL. Manthra connects automatically on next start.
MCP servers are stored in ~/.manthra/config.json under the mcpServers key. You can edit it directly or use the web UI.
id (unique string) · name (display name) · enabled (bool) · transport stdio | http · command + args + env for stdio · url for http
Tutorial
Give Manthra a real browser — navigate pages, click elements, fill forms, take screenshots, and scrape content.
Launch the built-in configuration panel from any Manthra session or directly from your shell.
Click the MCP Servers tab in the top navigation of the web UI, then click Add MCP Server.
Set transport to stdio. The npx command downloads and runs the Playwright MCP server on first use — no separate install needed.
Click Test in the web UI. On first run, npx will download @playwright/mcp and Playwright's browser binaries. Subsequent starts are instant from cache. A successful test shows a list of available tools.
npm install -g @playwright/mcp && npx playwright install chromium
Click Save. Restart Manthra — you'll see the connection status at startup:
By default Playwright MCP runs headless — the browser is invisible. Pass --headed to make it visible, which is useful for debugging or watching automations run live.
manthra web → MCP Servers → edit the server → update the Args field → Save, then restart Manthra.
Ask Manthra to use the browser naturally. It will automatically invoke Playwright tools as needed.
| Server | Package | What it adds |
|---|---|---|
| Playwright | @playwright/mcp@latest |
Browser automation — navigate, click, type, screenshot, scrape. |
| Context7 | @upstash/context7-mcp |
Live library documentation lookup — always up-to-date docs for any npm package. |
| Sequential Thinking | @modelcontextprotocol/server-sequential-thinking |
Step-by-step reasoning scaffold for complex multi-part problems. |
| GitHub | @modelcontextprotocol/server-github |
Search repos, read files, create issues and PRs via GitHub API. Requires GITHUB_PERSONAL_ACCESS_TOKEN. |
| GitLab | @zereight/mcp-gitlab |
GitLab repos, merge requests, and issues. Requires GITLAB_PERSONAL_ACCESS_TOKEN + GITLAB_API_URL. |
stdio, command to npx, args to -y <package-name>, and any required API keys in the env fields.
Features
No cloud sync, no IDE plugin. Just a fast terminal that does the work.
Use Ollama locally for fully offline operation, or connect to cloud providers — Groq, OpenRouter, Zen, Cerebras, or any OpenAI-compatible API.
Reads, writes, and edits files. Runs shell commands. Fetches URLs. Makes HTTP requests. The AI does real work, not just advice.
Drop an AGENTS.md in your repo with architecture notes, conventions, and build instructions. Manthra loads it automatically. Generate one with /init.
File operations inside your project directory are always allowed. Paths outside your working directory prompt for confirmation before proceeding.
Responses stream token by token with a pinned status bar and input line. Press ESC to cancel generation at any time.
Use /remember to save facts across sessions. Memory is automatically injected into the system prompt each time you start Manthra.
Run manthra web to open a browser-based panel for managing providers, models, and settings without editing config files by hand.
Reference images in your prompts with @path/to/image.png. Manthra encodes them and sends them to vision-capable models.
Toggle deep reasoning mode with /think. Supports off / low / medium / high budgets. Displayed in a collapsible thinking box.
Enable sub-agent spawning from manthra web. The model can delegate self-contained subtasks to a focused sub-agent with its own tool loop, then fold the result back into the main response.
Native app for Mac, Windows, and Linux. Full chat interface with multi-agent cards, file attachments, history sidebar, and a visual settings panel — no terminal required.
Connect Manthra to Discord as an always-on bot. Full tool access — reads files, runs commands, uses MCP tools. Permission prompts via emoji reactions. Configure and control from manthra web.
Channels
Connect Manthra to Discord as a persistent listener. It reads files, runs commands, and uses all your MCP tools — just like the CLI.
Go to the Discord Developer Portal, create an application, add a Bot, and copy the bot token.
Open manthra web → Channels tab → Add Channel. Select Discord, paste your bot token, and pick a provider + model.
Choose how the bot responds: only when @mentioned, on every message in specific channels, or triggered by keywords.
Click Start — the bot connects to Discord and begins listening. Monitor live status, message counts, and uptime from the web UI.
The Discord bot has the same tool access as the terminal — file I/O, shell commands, git, web search, and all configured MCP servers.
Track your bot in real-time from the web UI. See online/offline status, message counts, uptime, and errors at a glance.
Reference
Type any command at the Manthra prompt. All commands start with /.
| Command | Description |
|---|---|
| /help | List all available slash commands with descriptions. |
| /model <id> | Switch the active model mid-session. Run without arguments to see available models. |
| /clear | Reset the conversation history. The AI starts fresh with no memory of the current session. |
| /remember <text> | Save a note to persistent memory. It will be included in every future session's system prompt. |
| /forget <id> | Remove an entry from persistent memory by its ID (shown by /memory). |
| /memory | List all stored memory entries with their IDs and content. |
| /init | Generate an AGENTS.md file in the current directory. The AI scans your project and writes architecture notes and build instructions. |
| /web | Open the web configuration UI in your browser (runs on port 4875 by default). |
| /context /ctx | Show current context usage — message count, estimated token count, and percentage of the context window used. |
| /doctor | Run a diagnostic check on your provider and model configuration. Shows what's connected and what isn't. |
| /think [off|low|medium|high] | Toggle extended thinking mode. Set budget level or turn off. Thinking output is shown in a collapsible box. |
| /format [off|json] | Force the model to respond in JSON format. Useful for structured data extraction. |
| /exit | End the session. Conversation history is saved automatically to ~/.manthra/conversations/. |
Ctrl+E opens your $EDITOR for composing multi-line messages ·
Shift+Enter / Option+Enter adds a new line without submitting ·
Ctrl+V pastes from clipboard ·
ESC cancels an in-progress generation ·
Ctrl+D exits
Models
Use local Ollama models or connect to cloud providers. Switch models mid-session with /model — no restarts needed.
Compatibility
Single binary — no Node.js, no runtime, no dependencies required.
x64 and arm64. Works on Ubuntu, Debian, Fedora, Arch, and any modern distro with glibc ≥ 2.17.
Intel and Apple Silicon (M1 / M2 / M3). macOS 12 Monterey or later. The curl installer handles Gatekeeper automatically.
Via Git Bash or WSL2 (recommended), or as a native .exe. Windows Terminal gives the best rendering experience.
CLI or desktop. Local or cloud. Your choice.