A native desktop workspace that runs any coding agent against your own repositories. Connect the AI providers you already use today—or let Enigma handle model access through one subscription when Enigma-managed AI launches.
IDEAgent
◧◫◨
Implemented the workflow with order lookup, policy checks, and a human handoff for high-value refunds.
Order lookup and refund routing are connected. High-value refunds now require human approval.
| Scenario | Route | Result | Time |
|---|---|---|---|
| Order status | lookup_order | Passed | 340ms |
| Refund ≤ $100 | auto_refund | Passed | 510ms |
| Refund > $100 | human_handoff | Passed | 190ms |
Direct the work in a focused agent interface, with tool runs, changes, and outcomes kept visible.
Auto-presentingMost editors give you their agent. Enigma runs yours — Claude Code, Codex, Gemini CLI, Copilot and 33 more, hosted over the Agent Client Protocol as threads beside Enigma's own built-in agent. Same files, same approval gates, same window.
Bring your own provider without creating an Enigma account, or use an agent CLI you already have. Enigma-managed model access will add a subscription option when it launches.
Run the Windows installer, or unpack the Linux tarball and run ./install.sh. On Linux everything lands in your user folder — no sudo. On Windows the MSI installs to Program Files like any other app.
Connect one of 11 supported providers, pick an agent you already have installed, or join the waitlist for Enigma-managed AI with model access handled through one subscription.
Local, over SSH, or inside a WSL distro. The explorer, terminal, search and git all run wherever the code actually lives.
Choose a permission profile, type what you want, and watch every tool call, diff and shell command as it happens. Approve the risky ones, or set read-only and just ask.
Enigma's agent doesn't just start typing. Complex work gets a written plan, every risky action gets your approval, and the run isn't over until the change is checked.
A question gets an answer. A review gets findings by severity with path:line. Only an explicit edit request touches files — and for anything non-trivial the agent states its approach first so you can course-correct early.
Problem, blast radius, approach, a checklist and verification steps — written to .plan/PLAN.md in your workspace. When a choice is genuinely yours to make, the agent asks with a recommendation and one reason, never a blank menu.
Your permission profile decides what needs a yes. In the default profile every write, delete and shell command pauses for you: Enter to approve, or Reply instead to reject it and tell the agent what to do instead. Hosted agents route their own prompts through the same dialog.
The checklist pins to the bottom of the thread and ticks the moment each item lands — never batch-checked at the end. If the agent changes files and doesn't verify, the runtime itself reminds it to. Debugging follows a fixed order: prove the cause, one fix, confirm it's gone.
Hosted agents are declared in acp.json — a command, its arguments and any environment it needs. Pick one from the in-app registry or add your own. The same file drives launches on Windows, inside WSL, and over SSH.
Drop an AGENTS.md or CLAUDE.md at the repo root and attach it to a thread with one chip. Add MCP servers — local or remote — and the agent gets their tools.
Per-project memory carries what worked into the next thread. For high-risk decisions it can consult an advisor model — a second reviewer that sees the whole conversation.
Keep direct control with your existing provider accounts, or choose a simpler Enigma-managed subscription when it launches. The editor and agent workflow stay the same.
Connect supported AI providers with your own credentials. Choose the models you want and keep usage and billing directly with each provider.
Get access to supported AI models through Enigma without configuring separate provider credentials. Model access, usage, and billing will live in one place.
Your choice remains open. Enigma-managed AI is planned as an optional convenience, not a replacement for connecting your own providers.
When you bring your own provider, Enigma is a native desktop application with no Enigma account or relay in the request path. Your credentials remain on the machine, and model requests go directly to the provider you configured.
In bring-your-own-provider mode, requests go from your machine straight to the provider you configured. Enigma-managed AI will use a separate managed service path, with its data terms published before subscriptions open.
Anthropic, OpenAI, DeepSeek, Moonshot, MiniMax, Z.ai, xAI, Xiaomi MiMo,
OpenRouter, OpenCode and Cursor. Keys live in ~/.enigma/providers.json, never
compiled into the binary or sent to Enigma. Enigma-managed AI will not require
separate provider credentials.
Per-thread permission profiles — read-only, default, or full — and an approval prompt before any write, delete or shell command. Agents get autonomy inside a boundary you set.
SSH and SFTP workspaces, and WSL distros mounted as real Linux roots — the editor, terminal, search and agent all run on the far side, not on a synced copy.
Written in Rust on a GPU-accelerated UI. A 46 MB binary that starts cold in under a second and holds a large repository open without a fan spinning up.
Only what you send to the model provider you chose—your prompt and the files the agent reads—under that provider's terms. In BYOK mode, Enigma does not relay or bill those model requests.
The one outbound call the app makes on its own is a once-a-day version check to enigma.studiocoder.com/latest.json. That request carries a user-agent string and your IP address, and nothing else.
If you bring your own provider, credentials are stored in ~/.enigma/providers.json on your machine, readable only by your user. They are never compiled into the binary or sent to Enigma. Treat that file as you would an SSH key: disk encryption on, and don't commit your home directory.
No. You can connect supported AI providers with your own credentials and keep billing directly with them. Enigma-managed AI will be an optional subscription when it becomes available.
It is an upcoming subscription that will provide access to supported AI models without requiring separate provider credentials. Model access, usage, and billing will be managed through Enigma.
The final model catalog, usage limits, pricing, and managed-service data terms will be published before subscriptions open.
Yes. Bring-your-own-provider access is intended to remain available. Enigma-managed AI is a simpler alternative, not a replacement for direct provider access.
That is the planned experience: choose the connection that fits your workflow inside Enigma. Final switching behavior and subscription rules will be documented before launch.
The editor, terminal, search and git work fully offline. Agents need to reach whichever model you point them at — a local model through Ollama keeps that on-machine too.
Outbound requests honour the standard HTTP_PROXY, HTTPS_PROXY and NO_PROXY environment variables.
Enigma is free to use with supported providers you connect yourself. In that mode, you pay providers directly at their rates and Enigma does not sit in the billing path.
Enigma-managed AI will be a separate optional subscription. Pricing, included models, and usage limits have not been announced; they will be shown clearly before purchase.
No. Enigma is proprietary software, free of charge for use under its terms. The source is not public. You can inspect exactly what it does on the network with any packet capture tool — there's very little to see.
Any agent that speaks the Agent Client Protocol — 37 are bundled in the registry, including Claude Code, Codex CLI, Gemini CLI, GitHub Copilot, Qwen Code and Grok Build — and you can add your own with one entry in acp.json.
Enigma's own built-in agent currently accepts credentials for 11 providers: Anthropic, OpenAI, DeepSeek, Moonshot, MiniMax, Z.ai, xAI, Xiaomi MiMo, OpenRouter, OpenCode and Cursor.
The model catalog for Enigma-managed AI has not been announced. Supported models will be published before the subscription launches.
Every thread has a permission profile. Read-only lets the agent look but not touch. Default asks you before any write, delete, move or shell command. Full is for when you've decided to trust it. Agents hosted over ACP route their own permission prompts through the same approval dialog, so the boundary is consistent whichever agent is driving.
Windows 10 and 11, x64, via the MSI installer. Linux x86_64 with glibc 2.39 or newer — Ubuntu 24.04, Debian 13, Fedora 40 and later — via the tarball. macOS is in development; leave your email to be told when it ships.
Enigma checks for a new version once a day and offers to install it. Every release is also listed on the releases page with its checksum, so you can verify a download before you trust it.
Get launch details for managed model access, pricing, and supported models—plus the releases you choose below.