Snipara - Project Memory for AI Agents

Snipara gives AI coding agents a project-scoped memory and context layer that survives sessions, users, tools, and model switches. Your agent keeps using its own LLM; Snipara supplies durable decisions, source-backed retrieval, shared context, code graph context, and team memory so the next agent does not start cold.

Current extension release: 2.0.4.

Available from the VS Code Marketplace and Open VSX Registry for VS Code, Cursor, VSCodium, and other VS Code-compatible editors.

What This Extension Does

• Runs a no-account demo against Snipara's project context. The demo is opt-in and limited to 3 queries.
• Connects VS Code-compatible editors to a Snipara project through API credentials stored in VS Code SecretStorage.
• Lets users query source-backed project context, upload documentation, inspect memories, review decisions, and monitor index health.
• Exposes 63 Command Palette actions across project context, memory, documents, orchestration, sandbox execution, swarm coordination, and local readiness.
• Exposes 38 GitHub Copilot Language Model Tools in VS Code agent mode.
• Registers a Snipara MCP Server Definition Provider for VS Code's MCP integration when the host supports it.
• Keeps Cursor users supported through the extension UI plus the separate hosted MCP setup for Cursor chat.

Quick Start

No sign-in is required to try the demo.

1. Install Snipara from VS Code Marketplace or Open VSX.
2. On first unconfigured launch, the extension opens the Getting Started walkthrough once.
3. Click Try Demo Query to run the demo. The extension no longer auto-runs the demo on activation.
4. Click Sign in with GitHub to create or connect a free Snipara account.
5. Sync workspace documentation and start asking questions against your own project context.

Editor Support

Cursor and other Open VSX editors do not expose VS Code's GitHub Copilot-specific Language Model Tool APIs. For AI chat access in Cursor, configure the hosted MCP integration from Snipara Cursor docs.

Core Features

Demo and Onboarding

• Opt-in demo mode: Run 3 demo queries against Snipara's own indexed project context without signing in.
• Instant first demo: The default demo query uses embedded fallback data, so it works even when the network is unavailable.
• Live follow-ups: Custom demo questions call the read-only demo project and fall back silently if the API is unreachable.
• Guided walkthrough: First-run users see the Getting Started flow, but the demo only runs after the user clicks Try Demo Query.
• Workspace detection: Markdown workspaces are scanned and surfaced in the Welcome view so users can index project docs after sign-in.

Project Context

• Natural-language project context queries with keyword, semantic, or hybrid search.
• Source-backed retrieval with cited files, sections, relevance scores, and token budgets.
• Raw chunk, raw document, line-range, and project-load commands for deeper inspection.
• Multi-query, multi-project query, query decomposition, planning, and orchestration workflows.
• Code graph lookup for neighbors, callers, imports, importers, and structural paths.

Agent Memory

• Store and recall facts, decisions, learnings, preferences, todos, and context.
• Use Remember If Novel to avoid duplicate durable memories.
• Run End-of-Task Commit to persist reusable outcomes at the end of a task.
• Browse memories in the sidebar and inspect memory health, review queues, stale entries, and candidate memories.
• Query structured project decisions directly.

Project Intelligence

• Index health, recommendations, reindexing, search analytics, and query trends.
• Shared context collections and prompt templates for team standards.
• Summary storage for generated document summaries.
• Swarm coordination commands for multi-agent work: swarms, tasks, resource claims, shared state, and broadcast events.

Sandbox and Local Readiness

• Native Doctor / Local Readiness view for workspace .env files, provider keys, Snipara auth, Snipara Sandbox, and Docker.
• Sandbox commands for Docker-isolated execution, local execution, logs, and trajectory visualization.
• Local setup guidance installs the PyPI package snipara-sandbox.
• Core extension features do not require snipara-companion.

VS Code Integration

• Snipara activity bar with Welcome, Results, Session Context, Memories, Doctor / Local Readiness, and Swarm Dashboard views.
• Status bar entry for sign-in or quick query.
• Cmd+Shift+R / Ctrl+Shift+R shortcut for Snipara: Ask Question.
• File explorer upload for .md, .mdx, and .txt files.
• Optional auto-restore and auto-save of session context.

Commands

All 63 commands are available from the Command Palette under Snipara, Snipara Sandbox, and Snipara Doctor.

Query and Context

Memory and Decisions

Index, Documents, and Shared Context

Raw Context and Code Graph

Swarm Coordination

Sandbox, Doctor, and Setup

Copilot and MCP

In VS Code agent mode, Snipara contributes 38 Language Model Tools covering:

• project context retrieval, search, quick ask, multi-query, planning, decomposition, and multi-project search;
• memory storage, recall, browsing, deletion, novelty checks, memory health, memory review, and end-of-task commits;
• shared context, document upload, raw document load, full project load, orchestration, and REPL context packaging;
• source chunk loading, code graph neighbors, callers, imports, and shortest paths;
• structured decisions, index health, recommendations, reindexing, search analytics, and query trends;
• hierarchical task inspection, recommendations, metrics, and Python execution through Snipara Sandbox.

The extension also declares a Snipara MCP Server Definition Provider for VS Code hosts that support MCP server definitions. Hosted MCP availability depends on the editor, VS Code version, project configuration, and Snipara server deployment.

Settings

Requirements

• VS Code 1.93.0+, Cursor 0.40+, VSCodium, or another compatible Open VSX editor.
• No account is required for the opt-in demo.
• A free Snipara account is required to query and manage your own project context.
• Optional for sandbox execution: snipara-sandbox and Docker for isolated runs.
• Optional for companion workflows: snipara-companion.

Development

pnpm install
pnpm run type-check
pnpm run compile
pnpm run package

Packaging uses vsce package --no-dependencies. Publishing targets:

pnpm run publish:vsce
pnpm run publish:ovsx

Generated .vsix files are local release artifacts and are ignored by git.

Links

• VS Code Marketplace
• Open VSX Registry
• VS Code integration docs
• Cursor integration docs
• Dashboard
• GitHub
• Issues