Eymia AI Community

Eymia is local AI coding that finishes the job. Use local and open models to inspect, edit, run, test, repair, and verify code inside VS Code instead of only chatting about it.

Community is free for one local workspace. It includes the complete approval-based local coding loop with diff review, basic rollback, local model backends, and no Eymia token charges.

What Is Eymia?

Eymia wraps local and open models in a host-controlled development workflow. The extension gathers project context, lets the model propose focused actions, applies safety checks, asks for approval before risky work, runs verification, and reports the evidence.

That means your model can do more than explain code. It can help complete small real development tasks while VS Code keeps control over files, commands, approvals, rollback, and completion evidence.

Why Eymia?

Local models are useful, private, and increasingly capable, but coding-agent work fails for reasons that are not only about model intelligence. Agents can miss context, choose the wrong path, emit malformed tool calls, repeat failed edits, or claim success before tests pass.

Eymia adds the execution and verification loop around the model:

Understand -> Inspect -> Plan -> Edit -> Run -> Verify -> Repair -> Report Evidence

What Community Can Do

• Use local providers including Ollama, LM Studio, llama.cpp-compatible endpoints, and local OpenAI-compatible endpoints.
• Inspect the active editor, selected code, project files, diagnostics, Git status, and workspace structure.
• Search and read files inside one workspace root.
• Propose and apply file edits after approval.
• Show diffs and preserve basic rollback snapshots for accepted changes.
• Run approved commands and tests through the normal safety policy.
• Read command/test output, detect failures, attempt bounded repair, and rerun verification.
• Export conversations and report issues without sending source code to Eymia.

Community does not include remote/cloud model backends, advanced workspace indexing, MCP/A2A integrations, Git writes, VEX robotics, Autonomous mode, Multi Agent mode, shared team workflows, or enterprise governance.

60-Second Setup

1. Install and start Ollama, LM Studio, or another supported local model server.
2. Pull or load a coding model such as a Qwen Coder, DeepSeek Coder, or similar local code model that fits your machine.
3. Install Eymia AI Community in VS Code.
4. Run Eymia AI Offline Editor: Select Backend.
5. Run Eymia AI Offline Editor: Test Backend Connection.
6. Open a project folder.
7. Ask Eymia: Run the tests and explain any failures or Fix the current error and verify it.

Example Workflow

User: Fix the failing tests in this project.

Eymia:
Inspects the project.
Finds the relevant source and test files.
Proposes a file edit.
Shows the change for approval.
Runs the test command with approval.
Uses failing output as repair evidence if needed.
Reruns verification.
Reports files changed and the exact result.

Safety And Privacy

Eymia Community can run entirely with local models. Your source code is not sent to Eymia. If you configure a non-local provider in another edition, requests go to that provider according to your configuration and that provider's policies.

Community safety controls include:

• Explicit approval before risky file changes and command execution.
• Workspace-bound file access with path validation.
• Diff review before applying proposed patch bundles.
• Basic rollback snapshots for accepted workspace changes.
• Secret redaction in support bundles, task summaries, persisted state, and conversation export.
• Local privacy mode for stricter local-only behavior.
• Bounded steps, command limits, repair limits, and no-progress detection.

Eymia does not operate a source-code telemetry or file-ingestion service for this extension. See PRIVACY.md for the full privacy statement.

Community Vs Paid Editions

Supported Providers

Community is intended for local execution. Supported local backend families include:

• Ollama
• LM Studio
• llama.cpp-compatible OpenAI-style servers
• local Hugging Face-compatible endpoints when configured
• generic local OpenAI-compatible endpoints

Remote/cloud providers are paid-edition capabilities.

Package The Extension

For local release validation:

npm install
npm run test:unit
npm run package:edition:community

For the full stable Community release path:

npm run release:build:community

The stable Marketplace package should install from the normal Install button. The pre-release beta workflow remains available for internal validation with npm run release:build:community:beta.

Main Commands

• Eymia AI Offline Editor: Welcome
• Eymia AI Offline Editor: Select Backend
• Eymia AI Offline Editor: Test Backend Connection
• Eymia AI Offline Editor: Start Agent Task
• Eymia AI Offline Editor: Show Task History
• Eymia AI Offline Editor: Show Patch Preview
• Eymia AI Offline Editor: Apply Latest Patch
• Eymia AI Offline Editor: Restore Latest Rollback
• Eymia AI Offline Editor: Show Rollback History
• Eymia AI Offline Editor: Run Doctor
• Eymia AI Offline Editor: Export Conversations
• Eymia AI Offline Editor: Report a Bug

Documentation

• PRIVACY.md
• CHANGELOG.md
• RELEASE_QA_CHECKLIST.md
• docs/COMMUNITY_0.9.1_VALIDATION_PLAN.md
• docs/ARCHITECTURE.md
• docs/SAFETY.md
• docs/MCP.md
• docs/A2A.md

Report An Issue

Use Eymia AI Offline Editor: Report a Bug from VS Code. The report opens locally for review before you send anything. Privacy-safe support bundles exclude prompts, file contents, workspace paths, backend URLs, credentials, and environment variables.

Known Limitations

• Model quality depends on the local model and available hardware.
• Community supports one workspace root at a time.
• Remote/cloud model backends are paid-edition capabilities.
• MCP, A2A, Autonomous mode, and Multi Agent mode are paid-edition capabilities.
• Eymia can verify only with commands or tests available in the project.
• Some extension-host smoke tests require a local VS Code installation and may not run in headless or restricted environments.