LOKI AI Agent

LOKI is an AI coding agent for VS Code that can understand your codebase, investigate problems, make changes, run tests, and work through development tasks directly inside VS Code.

It is designed around two principles: give the agent enough context to actually work on a codebase, while keeping the developer in control of what it does.

What LOKI Agent Can Do

LOKI can work with your codebase as an actual coding agent rather than only answering questions about code.

It can:

• Understand and search your codebase.
• Read and analyze files and project context.
• Create and modify files.
• Fix bugs and diagnostics.
• Implement features and changes.
• Write and update tests.
• Run builds, tests, and terminal commands.
• Investigate failures and debug problems.
• Review code and changes.
• Refactor existing code.
• Explain how code works.
• Generate and improve documentation.
• Remember important project context across conversations.

Two Operating Modes

LOKI provides two ways to control how the agent works.

User Mode

You stay in control of every step. In User Mode, LOKI does not automatically continue through a multi-step task. Before performing the next significant action, it asks for your permission.

This mode is useful when you want to inspect and approve the agent's actions as it works.

Auto Mode

LOKI works through the task on its own. In Auto Mode, the agent can decide what actions are needed, execute them, inspect the results, and continue iterating until the requested task is completed.

Auto Mode is intended for tasks where you want LOKI to take care of the development workflow rather than approving every individual step.

Persistent Memory

Memory is a core part of LOKI Agent. LOKI can maintain persistent, encrypted memory across conversations so that useful information does not have to be rediscovered every time.

Memory can contain things such as:

• Important project conventions.
• Architectural decisions.
• Development preferences.
• Relevant project facts.
• Previous conversation context.
• Other information explicitly worth remembering.

🔒 Memory Infrastructure

• Memory is stored on your disk in encrypted form.
• The memory system is designed so that persistent information is not simply stored as ordinary plaintext Markdown files.
• LOKI can read and update its memory while working, allowing future conversations to start with relevant context already available.
• Security and control are central to LOKI's design.

AI Providers

LOKI is provider-agnostic and can work with multiple AI providers. This means you can use LOKI with hosted models or a local model without any worries.

Built for Real Development Work

LOKI is designed to handle tasks that require multiple actions rather than simply generating a code snippet.

A typical task might look like:

"Add support for dark mode and make sure the existing tests continue to pass."

LOKI can systematically:

1. Explore the relevant parts of the codebase.
2. Understand the existing implementation.
3. Identify the files that need to change.
4. Make the required modifications.
5. Add or update tests.
6. Run the tests.
7. Investigate failures.
8. Iterate on the implementation.
9. Present the resulting changes for review.

The same workflow can be performed interactively in User Mode or autonomously in Auto Mode.

Developer Control

LOKI is designed to give you a choice between control and autonomy. User Mode lets you approve the agent's actions as it progresses. Auto Mode lets the agent take ownership of the workflow and continue working without asking for more permissions at every step.

In both modes, the goal remains identical: Give the agent the ability to do real development work while keeping your code, credentials, and persistent context protected.