What is Snipara?
Snipara is a context and memory layer for AI coding agents. It gives each repository a Project Brain powered by Project Intelligence, connecting decisions, active work, code impact, verification evidence and handoffs across Claude Code, Cursor, Codex, Gemini CLI and other MCP clients. Multiple agents and humans work from the same project reality because Snipara keeps ownership, verification, handoffs, receipts and outcomes visible.
The simple frame
Talk to your AI agent like it is your lead engineer. Snipara gives the repository a Project Brain: the system of record for decisions, active work, code impact, verification, ownership, and outcomes. Project Intelligence is the technology that makes that brain work, and Companion is the optional local workflow layer that helps a model use it — start-work briefs, phase commits, handoffs, and resume points.
Category-wise, Snipara answers queries like "AI coding agent context," "persistent project context for Cursor," and "Claude Code project context." But the differentiator over a plain context tool is broader project state: not just recalled notes, but reviewed decisions, source-backed docs, active work, Code Graph impact, proof receipts, handoffs, and outcomes kept together.
What Snipara is not
- Not an LLM provider — Snipara does not sell inference. You bring your own model and client.
- Not an IDE or a fork of one — it connects into Claude Code, Cursor, Codex, VS Code, and other clients through hosted MCP.
- Not a chat product — the product is project state (decisions, active work, impact, evidence), not a conversation window.
- Not autonomous — Project Intelligence briefs are reviewable evidence, not automatic approval to merge or ship.
The architecture in one minute
An AI client connects through Hosted MCP or the API, which returns Project Intelligence. Companion, Orchestrator, and Sandbox are optional layers added only when a workflow needs local continuity, proof gates, multi-agent coordination, or execution evidence.
| Layer | Role | Required? |
|---|---|---|
| AI client | Claude Code, Cursor, Codex, ChatGPT, or another MCP/API-aware client — it still owns reasoning and execution. | Yes |
| Hosted MCP / API | The stable connection into Snipara. Agents use hosted MCP; other systems can use the API or SDK. | Yes |
| Project Intelligence | The project-owned layer that compiles context, reviewed memory, code impact, active work, decisions, and verification guidance. | Yes |
| Companion | Local workflow runtime for start-work briefs, phase commits, handoffs, resume points, and impact gates. | Optional |
| Orchestrator | A separate advanced package for production validation, drift checks, and explicit multi-agent coordination. | Optional |
| Sandbox | Optional execution support for validation, inspection, and repeatable proof — useful when a task needs to run code, not for every lookup. | Optional |
Who Snipara is for
How to start
The recommended path is one command. It asks which assistant or MCP client the project uses, writes hosted MCP references and client-specific helper files, and keeps merge-ready templates for any existing AGENTS.md or CLAUDE.md instead of overwriting them.
Limitations
- Snipara does not replace the coding model — it does not write your code.
- It does not guarantee correct code and does not remove the need to run tests.
- Project Intelligence briefs are reviewable evidence views, not raw transcript storage, causal proof, or autonomous approval.
- Snipara does not install hooks for every agent — hook support is conservative and client-dependent.
- Snipara is not an LLM provider; it does not sell inference.
FAQ
What is Snipara in one sentence?
Snipara is a context and memory layer for AI coding agents. It gives each repository a Project Brain powered by Project Intelligence, connecting decisions, active work, code impact, verification evidence and handoffs across Claude Code, Cursor, Codex, Gemini CLI and other MCP clients.
Does Snipara replace Claude Code, Cursor, or Codex?
No. Those tools still do the reasoning and the editing. Snipara provides the project context, reviewed memory, code impact analysis, and workflow continuity those tools use before and during the work.
Does Snipara sell access to an LLM?
No. Snipara does not sell LLM inference. Customers bring their own LLM or AI coding client and connect it to Snipara through hosted MCP or the API.
What is Project Intelligence?
It is the technology behind the Project Brain: a five-question model that tries to answer what changed, why it changed, what it impacts, how it should be verified, and whether it is safe to proceed, built from context, reviewed memory, code graph structure, and workflow state.
How do I start using Snipara?
Run npx create-snipara. It asks which assistant or MCP client you use, writes hosted MCP references and client-specific helper files, and keeps merge-ready templates for any existing AGENTS.md or CLAUDE.md instead of overwriting them.
Is Companion required to use Snipara?
No. Companion, Orchestrator, and Sandbox are optional layers added when a workflow needs local continuity, proof gates, multi-agent coordination, or execution evidence. A basic setup only needs hosted MCP.