AtlasMind

_{· Current source version: 0.127.2 ·}

BETA
AtlasMind is your AI teammate for solo and small dev teams.
Ship faster, automate the boring parts, and keep your project's brain in one place — all inside VS Code.

AtlasMind is built for indie developers, freelancers, and small teams who want to get more done without context switching or tool overload. It's not just a chatbot — it's a multi-agent orchestrator that routes your tasks to the right AI, remembers your decisions, and helps you focus on what matters most.

Why solo and small devs love AtlasMind:

• No more context switching: Everything happens in your editor — chat, code, memory, and planning.
• Automate the grind: Refactoring, testing, docs, and more — handled by specialized agents.
• Bring your own models: Use Local LLM, OpenAI, Claude, Gemini, Azure, or your favorite provider. Mix and match for cost, speed, or quality.
• Project memory that sticks: AtlasMind remembers your architecture, decisions, and lessons learned, so you don't have to.
• Stay in control: Approvals, cost tracking, and safety guardrails keep you in the driver's seat.
• Secure and reliable by default: Strong security guardrails and a configurable 23-methodology testing strategy system — TDD, BDD, E2E, security, performance, and more — with per-agent assignment, auto-detect, and AI token impact guidance so you can build with confidence from day one.
• Everything at a glance: Project, run, personality, and cost dashboards keep you in control — review agent runs, memory, and spend in one place.

What Makes AtlasMind Different?

• Multi-agent orchestration: 15 built-in specialized agents — debugger, frontend/backend engineers, reviewer, security, SEO, UX, DevOps, and more — plus instant AI-drafted custom agents on demand.
• Multi-provider model routing: Supports GitHub Copilot, Claude, GPT, Gemini, Azure OpenAI, Bedrock, Mistral, and more. Budget and speed preferences steer selection automatically.
• Built-in skills: 43 pre-built skills including file editing, git, diagnostics, code navigation, test running, debugging, HTTP requests, Docker, web fetch, and more. Skills are grouped by category and support custom folders. Agents use AI-driven auto skill assignment by default.
• Long-term project memory (SSOT): Decisions, architecture notes, and lessons learned persist in a structured memory folder. A dedicated Memory Agent maintains session context and keeps SSOT snippets fresh as source files evolve.
• Project planner: Decompose goals into subtasks, preview impact, gate execution, and review results.
• Cost tracking: Real-time per-session spend with budget guardrails and a daily cost limit.
• MCP server support: Extend AtlasMind with Model Context Protocol (MCP) servers for custom tools, agent extensions, and advanced workflows.
• Voice & Vision: Speak your prompts and hear responses via the Voice Panel (TTS/STT). Attach workspace images to any question via the Vision Panel for multimodal analysis.
• Session management: Name, organize into folders, archive, and restore chat sessions for long-running projects.

Quick Start

1. Install AtlasMind from the VS Code Marketplace.
2. Open AtlasMind: Manage Model Providers and configure your first model provider.
3. Start AtlasMind in your workspace:

• For a new project, run @atlas /bootstrap.
• For an existing project, run @atlas /import.

4. Ask AtlasMind to help with your next task.

For advanced setup, provider notes, CLI usage, or development workflows, see:

• Getting Started
• CLI Usage
• Model Routing
• Development Guide

Focused provider test example:

• npm run test:providers:local-recommendations (validates local-model recommendation override loading and fallback behavior; this also runs as an explicit CI quality gate)

Chat Slash Commands

Use these in the AtlasMind chat panel by typing @atlas /<command>.

Extension Commands

Access these from the VS Code Command Palette (Ctrl+Shift+P).

See Remote Control for the architecture and security model.

Sidebar Views

AtlasMind adds a sidebar with the following tree and webview panels:

Built-in Agents

AtlasMind ships 15 specialized agents, automatically routed by task type.

Agents use AI-driven skill auto-assignment by default — AtlasMind selects the best-fit skills for each agent's role automatically. Skills can also be assigned manually per agent.

Agents can be auto-updated on a configurable cadence (never/daily/weekly/monthly/every-use) so system prompts stay current with best practices and compliance requirements. Individual agents can opt out of auto-updates.

Built-in Skills

43 built-in skills organized by category. All skills are enable/disable toggleable and undergo security scanning before use.

Custom skills can be authored and loaded from any workspace folder. The Skills view supports folder organization and per-skill security scanning.

Configuration

Key settings under atlasmind.* in VS Code settings:

See Configuration Reference and wiki/Configuration.md for the full settings list.

Open Source & Support

AtlasMind is fully open source and available under the permissive MIT license. There are no paywalls, feature gates, or commercial editions—just the full project, free for everyone.

If AtlasMind saves you time or helps your team, consider a pay-what-it's-worth donation to keep the project alive and thriving. Every bit of support helps sustain ongoing development.

See Funding and Sponsorship for details.

Learn More

• Core Workflows
• Model Routing
• Agents & Skills
• SSOT Memory System
• Configuration Reference
• Roadmap
• Comparison Matrix
• Funding and Sponsorship

Project Structure

• Core runtime: src/core/, src/runtime/, src/chat/, src/commands.ts, src/extension.ts
• Provider adapters and catalogs: src/providers/ (including localModelSync.ts and localModelRecommendationRegistry.ts)
• Skills and tool handlers: src/skills/
• Shared utilities: src/utils/ (including secretRedactor.ts — pattern-based secret scanner used to scrub credentials from memory context before LLM dispatch; aiInstructionSync.ts — inbound merge of external agent rule files; testingProtocolSync.ts — outbound sync of enabled testing protocols into external agent instruction files; terminalOutput.ts — strips ANSI/control escape sequences from captured tool output before it is shown in chat summaries or webviews)
• Data privacy: src/core/dataPrivacyManager.ts (classifies confidential/proprietary terms, files, and folders and gates them to user-selected "trusted" models; records catch activity for the dashboard charts), src/core/compliancePacks.ts (built-in GDPR/HIPAA/PCI-DSS/CCPA detector packs), and src/core/providerDataGovernance.ts (per-provider GDPR/data-management reference links). Managed from the Project Dashboard → Privacy page (provider/model trust tree, catch charts, and provider data-management panel); policy stored at project_memory/operations/data-privacy.json.
• Delivery & deployment stages: src/core/deliveryManager.ts (models Local → Staging → Production stages and promotion "push" edges; seeds a pipeline from the repo's branches, sanitises dashboard edits via sanitizeDeliveryConfig, and persists project_memory/operations/delivery.json + a human-readable delivery.md runbook mirror) and src/core/promotionRunner.ts (the guarded promotion engine: builds the preflight → backup → deploy → verify → record plan, enforces the authorization gate, and executes user-authored commands with live progress). Surfaced on the Project Dashboard → Delivery page as an editable Stages & Promotion pipeline with Execute / Runbook push buttons; production is protected, a data-bearing target with no backup command is deny-by-default blocked, executed commands are sourced only from your saved config/routines, and AtlasMind never force-pushes. The page auto-refreshes on external delivery.json changes (file watcher) and shows a "review needed" banner when the protocol, stage-candidate branches, or CI workflows have drifted since your last review.
• Mission Loop (autonomous goal-seeking loop): src/core/missionRunner.ts (the outer plan → execute → evaluate loop with the closed parameter envelope and deny-by-default checkpoints), src/core/goalEvaluator.ts (validated, untrusted-output progress verdicts), and src/core/missionRegistry.ts (audit persistence to project_memory/operations/missions.json + a missions.md runbook mirror). Defined/launched/watched from the Mission Control webview (src/views/missionControlPanel.ts) and the /loop chat command.
• Testing strategy: src/core/testingConfigLoader.ts (methodology resolution for orchestrated runs) and src/core/testingScaffolder.ts (stack-aware framework scaffolding)
• Routing intelligence: src/core/executionQuality.ts (shared output-quality scorer), src/core/modelEvalHarness.ts (scored-replay model comparison), and src/views/modelComparisonPanel.ts (comparison webview)
• Webview and sidebar surfaces: src/views/ (chatProtocol.ts and chatWebviewMarkup.ts are Node-free so they are shared with the web build)
• Voice (TTS/STT): src/voice/ (voiceManager.ts bridge, hostSpeechSynthesizer.ts on-device OS speech engine, localTranscriber.ts on-device Whisper STT)
• Memory and MCP layers: src/memory/, src/mcp/
• Agentic Resource Discovery: src/ard/ (ardClient.ts protocol client, ardRegistry.ts Agent Finder registry, ardInstaller.ts install mapping, ardCatalogExporter.ts publisher). The discovery UI is the Resource Discovery tab in the Settings dashboard (src/views/settingsPanel.ts). See Resource Discovery.
• Remote control: src/remote/ (protocol.ts wire format, remoteControlServer.ts desktop server, remoteBridge.ts synthetic webview host) and src/web/ (browser thin-client entry, remoteClient.ts, chatClientPanel.ts, dashboardPanel.ts)

Contributing & License

See CONTRIBUTING.md for dev setup and contribution guidelines.

MIT License — see [LICENSE]