Sniparaproject intelligence
Menu
Product
Workflow loopConnect once, load project intelligence, carry state forward.Why SniparaThe explicit difference from an agent working alone.Project IntelligenceShared state, decisions, impact, outcomes, and verification before edits.Agent surfacesHosted MCP, Code Graph impact, Sandbox proof, and Companion workflows.Proof replayA controlled before/after replay on real Snipara engineering work.Evaluation methodHow Snipara measures discovery, planning quality, and verification evidence.Verification plansRecommended checks, risk, and missing gates before edits.Business ContextHelvabase-owned business workflows powered by Snipara context infrastructure.PlansOne tier includes context, intelligence, and agents.VisionWhy intelligence belongs to the project workspace.
Developers
QuickstartConnect an AI coding agent in minutes.Project Intelligence modelThe five-question model behind agent work.Hosted MCPTool contracts for agent clients.API ReferenceProgrammatic context and project intelligence access.GitHub AppRepository sync and PR Answer Packs.Workflow ContinuityBriefs, handoffs, impact, and review context across agent work.Snipara SandboxRepeatable execution when validation needs proof.Companion CLIWorkflow continuity for long agent work.
VisionProofPricingDocs
Sign inStart Free
Docs / Architecture

Snipara architecture in one minute.

Start with an AI client, connect it to Hosted MCP or API, then let Project Intelligence prepare the project state. Companion, Orchestrator, and Sandbox are optional layers you add when the workflow needs local continuity, proof gates, or execution evidence.

Snipara component map
Rendering diagram...
Required path

The smallest real Snipara setup has three layers.

A developer does not need to install every package to understand Snipara. The core contract is client, hosted boundary, and shared Project Intelligence.

AI client

Claude Code, Cursor, Codex, OpenAI Agents, ChatGPT, or another MCP/API-aware client still owns reasoning and execution.

Hosted MCP / API

The stable boundary into Snipara. Agents use Hosted MCP; software systems and integrators can use the API or SDK.

Project Intelligence

The project-owned layer that compiles context, reviewed memory, code impact, active work, decisions, outcomes, and verification guidance.

Optional depth

Add runtime layers only when the work earns them.

These components are deliberately visible, but they should not obscure the simple path for a new developer connecting an agent.

Companion

Local workflow runtime for briefs, phase commits, handoffs, resume, impact gates, and release checks. Useful for longer agent work, not required for a simple Hosted MCP install.

Orchestrator

Advanced Python package for proof gates, drift checks, htasks, and explicit multi-agent coordination. Install only when that control plane is needed.

Sandbox

Optional execution layer for repeatable validation, inspection, and proof evidence. It is useful when a task needs code execution, not for every documentation or retrieval task.

Interaction rules
  • Hosted MCP / API is the integration boundary; Snipara does not replace the user's model.
  • Project Intelligence is the core product layer. Memory, context, code graph, workflow state, and outcomes are inputs to it.
  • Companion and Orchestrator are separate on purpose: Companion is the local workflow helper; Orchestrator is the advanced validation and coordination package.
  • Sandbox sits below workflow tools when execution evidence is needed. Runtime targets such as Docker, local, or WASM belong to the Sandbox layer.
Start

Connect a client

Choose Hosted MCP for agents or API / SDK for software integrations.

Core

Understand Project Intelligence

See how Snipara answers what changed, why, impact, next step, and safety.

Packages

Pick optional layers

Compare create-snipara, Companion, Orchestrator, Sandbox, SDK, and MCP packages.