Alan AI Chat is a VS Code extension for building your own coding assistants. Create multiple agents, give each one its own instructions, choose the model it should use, and let it work with your workspace through explicit tools and guardrails.
Highlights
Agent templates: start from focused templates for code exploration, repository review, prompt polishing, translation, context loading, log trimming, and sensitive-data scrubbing.
Guard workflows: attach guards that run at session start, before a message is sent, before a tool runs, or after a tool returns. Guards can annotate, rewrite, block, coach, sanitize, or append system context.
Markdown definitions: agents, guards, and prompt fragments are stored as local Markdown files under .alan, so they can be reviewed, versioned, and shared with a project.
Safer tool use: configure tools per agent, require approval for risky actions, and auto-approve trusted read/search workflows.
Workspace-aware tools: agents can read files, write files, run commands, search with grep, activate skills, and delegate work to another agent.
Multiple model providers: use OpenAI, Claude, or local Ollama models from the same agent interface.
Better chat experience: improved streaming, clearer tool messages, cancellable runs, summaries, retry/continue flows, and dedicated views for guard execution details.
Appearance and streaming settings: choose Alan Light, Alan Dark, or follow VS Code, and tune how quickly streamed text appears.
What You Can Build
A code-review agent that annotates each request with relevant files before answering.
A lightweight coding agent with read, search, and write tools.
A prompt-polishing workflow that rewrites vague requests before the main model sees them.
A privacy-first workflow that scrubs sensitive data before it reaches the assistant.
A local-model workflow using Ollama for private or offline development.
A multi-agent setup where one agent can delegate focused work to another.
Local-First Configuration
Alan stores agent definitions, guard definitions, prompt fragments, and conversation data locally. API keys are stored with VS Code's Secret Storage API, which uses your operating system's secure credential storage.
Prompts and attached context are sent only to the model provider you choose for the active agent, such as OpenAI, Anthropic, or your local Ollama server.
Getting Started
Open the Alan AI Chat view from the VS Code activity bar.
Add your OpenAI or Claude API key in Settings, or refresh Ollama models after starting a local Ollama server.
Create an agent from a template, or start from a blank agent.
Open the agent definition to inspect or edit its tools, guards, fragments, model, and instructions.
Start a conversation and approve tool calls when Alan asks.
Agent Definitions
Agents are Markdown-backed definitions. A typical agent can include:
instructions for the assistant
the selected model and temperature
tools such as read_file, write_file, execute_command, and search_grep
guard bindings for pre-send, tool, and session-start workflows
prompt fragments and skills for reusable behavior
Because definitions live in the workspace, teams can review and evolve them alongside code.
Frequently Asked Questions
How much does it cost?
Alan AI Chat is free. You pay your selected model provider directly for API usage.
Which models can I use?
Alan supports OpenAI, Claude, and Ollama model sources. The available list depends on your API keys and the local Ollama models you have installed.
What happens with my data?
Alan stores its configuration locally. Your prompt, selected context, images, and tool results are sent to the active model provider so the assistant can answer. If you use Ollama, requests go to your local Ollama server.
Where do I get API keys?
Create keys from your provider account: OpenAI keys from the OpenAI dashboard and Claude keys from the Anthropic console.
Can I use Alan without cloud models?
Yes. Start an Ollama server, pull one or more models, then refresh Ollama models from Alan's Settings view.
