AI coding that lives in your IDE and can run fully offline.
Local LLMs via Ollama, LM Studio, or vLLM — or bring your own Anthropic, OpenAI, or Bedrock key. Six agent modes. Zero telemetry. Zero cloud sign-in.
Elixium Coder talks only to the endpoints you explicitly configure. No account required. No usage data collected. Designed for air-gapped environments, privacy-conscious teams, and anyone who thinks the IDE is still the right home for their coding agent.
Source-available, single-vendor. All code is public and auditable at https://github.com/IndirectTek/elixium-coder. Every commit is authored and signed by IndirectTek. No external contributions accepted — federal and enterprise customers get a provable single-vendor supply chain.
Migrating from Roo Code?
Roo Code is sunsetting on May 15, 2026, and the team is pivoting to cloud-only agents. If you want to stay in your IDE, keep your existing workflow, and never depend on a vendor's cloud being online — Elixium Coder is the landing spot.
The architecture and mode system will feel familiar. To bring your setup across:
- Install Elixium Coder from the Marketplace.
- Re-enter your API credentials in Settings → Providers (they're stored in VS Code's secret storage, not shared across extensions).
- Copy any workspace rule files from
.roo/rules/to.elixium/rules/(and.roo/rules-{mode}/to.elixium/rules-{mode}/). The format is the same. - Copy
.roomodescontents to.elixium/modes.yamland.roo/mcp.jsonto.elixium/mcp.jsonif you use those. - Custom modes from the Roo modes UI can be exported and re-imported through Settings → Modes.
No cloud migration needed because there's no cloud. Your API keys and conversation history never leave the machine.
Install
In VS Code: Extensions → search Elixium Coder → Install.
Or from the command line:
code --install-extension IndirectTek.elixium-coder
Providers
Five providers are supported. Pick one:
| Provider | Use for |
|---|---|
| OpenAI-Compatible | Ollama, LM Studio, vLLM, LocalAI, any OpenAI-compatible API |
| Anthropic | Claude Opus / Sonnet / Haiku |
| OpenAI | GPT via api.openai.com (also covers Azure OpenAI) |
| OpenAI-Native | OpenAI Responses API (reasoning models, stateful tool calls) |
| Bedrock | Claude / Llama / Titan via AWS Bedrock |
Credentials are stored in VS Code secret storage.
Modes
Six built-in agent modes:
- 🏗️ Architect — plan and design before implementation
- 💻 Code — write, modify, refactor
- ❓ Ask — answers and explanations, no code changes
- 🪲 Debug — diagnose and fix issues
- 🪃 Orchestrator — coordinate across modes for multi-step projects
- 🎨 Designer — visual UX design that outputs renderable HTML/Tailwind, React components, SVG sketches, and Mermaid flows
Switch modes at any time from the dropdown in the chat input.
MCP
MCP servers are configured via .elixium/mcp.json (workspace) or your
global config, and speak only to the endpoints you specify. See
https://modelcontextprotocol.io/ for server setup.
Network posture
Elixium Coder's only outbound traffic is to:
- Your configured LLM endpoint
- Your configured MCP servers
No telemetry. No marketplace callback. No sign-in. No background fetches to any vendor domain.
Verified clean in the shipped bundle: zero references to any cloud telemetry, auth, or routing service.
Support
- Website: https://elixium.ai
- Email: [email protected]
License
Apache License 2.0. Copyright 2026 IndirectTek. Derived from prior open-source work; see the NOTICE file inside the extension for attribution.