Skip to content

Project memoryand context foryour coding agents.

The second brain of your project, in Claude Code, Cursor and Codex. Find the source. Keep the learning. Check the change.

Project context for the tools you already use

Claude CodeCodexCursorVS CodeModel Context Protocol

From project knowledge
to a checked change.

Your model reasons and uses its tools. Snipara supplies project sources, durable memory and local workflow support.

  1. Context

    Load project sources, recalled memory and code impact.

  2. Plan

    Record the plan, its phases and scoped tasks.

  3. Work

    Make the scoped change with your existing coding client.

  4. Validate

    Run external checks and retain the evidence.

  5. Review

    A separate review session inspects the change; you choose who runs it.

  6. Decide

    With the Sandbox, the change carries its check receipts. Without it, Snipara says what was not checked. You decide.

MCP context access and native execution support are different. Native execution depends on the configured adapter, policy, model endpoint and permissions.

How the harness works

Three pillars. One project.

Retrieve sources, retain durable knowledge and keep local work tied to its checks. Keep your existing coding client and model.

Context: Ask with project knowledge. Hosted context uses your project account. Local work: Keep the work and its checks together. For deterministic local checks, no Snipara account is required. Compatible tools are prepared together; the three pillars do not require three separate setups. MCP context access and native execution support are different; execution depends on the client, adapter, policy and permissions.

Cloud context: Find the source passage.

Retrieve bounded passages from indexed project documents. Multilingual search combines vector and full-text ranks, with optional reranking and support judging.

Explore Cloud context

Cloud memory: Keep decisions with their evidence.

Recall durable knowledge across sessions. Source pointers, retirement, restore and outcome evidence keep each statement tied to its history.

Explore Cloud memory

Local coding harness: Continue the work and check the change.

Companion keeps workflow state, lead plans and receipts. Sandbox runs isolated checks and returns a receipt you can verify.

Explore Local coding harness

Your coding tools.
One Snipara plugin.

Install Snipara in Claude Code or Codex. Choose project Context for cited answers, or Local work for task continuity and checks without signup.

The plugin is free. Hosted Context uses your account and project access. Local work needs no Snipara account and prepares its compatible tools together. Standalone components remain advanced, optional integrations; Hosted service plans are separate.

Continue from a checkpoint.

Save the state of the work, resume the next phase and keep the review decision separate.

A saved-checkpoint example

Interactive illustration
Claude CodeEarlier work saved

Project checkpoint

Plan and decisions
Preserved
Completed tasks
Preserved
Attempts and evidence
Preserved
Next phase
Ready to resume
A fresh Codex sessionReady for the handoff
Your work has a place to return to.

Save the plan, completed tasks, attempts and evidence before the next session.

Illustrates the workflow, not a live run. Resume requires the project folder, saved artifacts and the project’s Snipara MCP.

The agent can change.
The work continues.

In one local demonstration, Claude Code completed the first two phases. A fresh Codex session resumed the saved checkpoint and continued.

Inspect the handoff evidence
Plan and decisions
Preserved
Original evidence
Preserved
Validation
Outside the coding session

10 October 2026. Operator-reported local proof.

What this demonstrates

One shared workspace and one fixture. Resume used local .snipara state, project files, referenced artifacts and the project’s hosted Snipara MCP. This does not establish live-process migration, MCP-only recovery, general reliability or time and cost savings.

Read the public receipt

Measured on the
running service.

Questions with a known answer, asked through the public client against the production service. The first passage returned is graded against the reference answer; questions that the documents cannot answer must come back empty.

Read the method and the limits
First passage answers the question
55 of 60
Unanswerable questions left empty
18 of 20
Useful memory returned before a task
14 of 19

13 November 2026. Snipara’s own project, measured by Snipara.

What this does not show

One project (Snipara’s own documentation and memories), question sets written by the team, answers graded by a language model. It is not an independent benchmark and says nothing about another codebase or about the quality of the code an agent writes.

Before you connect.

How do I install Snipara in Claude Code or Codex?

Install the free native plugin from Snipara’s publisher marketplace, connect its MCP server with OAuth and choose your Snipara project. Snipara, Companion and Orchestrator are enabled by default and can be switched off independently. Local engines require explicit setup and do not start automatically. A Snipara account is required; optional service plans are purchased separately. Official directory submissions are in preparation. Installation instructions are at snipara.com/plugins.

Can Claude Code work be resumed by Codex?

Companion can resume saved workflow state when the next coding tool has access to the project folder with its saved .snipara state and artifacts, and the project’s Snipara MCP. One local Claude Code to Codex test preserved the plan, completed tasks, attempt counters and evidence as work continued. This historical demonstration does not establish the behavior of the replacement Orchestrator runner. This is checkpoint continuity, not migration of a live process. MCP-only restoration and Orchestrator controller transfer were not tested.

What is Snipara?

Snipara provides cloud context retrieval, cloud memory and a local coding harness: Companion continuity, Sandbox testing and Orchestrator plan preparation with preset roles. Orchestrator lots 4–8 are not shipped.

Is Snipara a second brain?

For a project, not for a person. A personal second brain keeps your own notes. Snipara keeps what a project knows where every coding agent can reach it: source passages returned with their file and lines, and decisions kept with their evidence. It does not take notes for you and it does not replace your model.

What is a coding agent harness?

A coding agent harness is the system around a model that supplies context, scopes its work, manages execution and checks changes. The model proposes code; the harness defines the workflow and the conditions for accepting it.

Does Snipara replace Claude Code, Codex, Cursor, or ChatGPT?

Your model and coding client still reason and use tools. Snipara supplies project knowledge and workflow controls. MCP context access and native execution support are different capabilities; native execution depends on the configured adapter.

How is Snipara different from a context tool?

Context and memory ground the work. Companion tracks phases; Sandbox runs isolated checks. Orchestrator prepares role-bound plans, while lots 4–8 of its replacement runner remain unshipped. A passing test alone does not approve a change.

Which configurations have been tested?

An isolated Claude CLI journey reached acceptance with a test-local launcher. Qwen3 Coder Next through LM Studio passed external validation, then independent review refused a defect and blocked integration. These tests use one exposed development fixture and do not establish general model reliability.

Your models.
Your project’s second brain.

Connect the tools you already use.
Give the project a memory that stays.

Connect your coding agent