Savyre Code Intelligence

AI coding assessment, workflow, and evaluation inside your IDE.

Savyre Code Intelligence is a privacy-first IDE extension that helps teams assess, guide, and evaluate AI-assisted coding. It works alongside tools such as GitHub Copilot, Cursor, Codex-style agents, Claude Code, and other AI coding assistants.

Savyre is not another code generator. It is a structured workflow and evaluation layer that helps developers understand requirements, use AI responsibly, review generated code, run validation, debug issues, consider risks, and take final ownership.

Why Savyre?

AI can generate code quickly. The harder question is whether the developer can explain it, validate it, test it, debug it, and own it.

Savyre helps teams bring structure, evidence, and accountability to AI-assisted coding by guiding developers from task understanding to final review.

Use Savyre when you want to:

• Run realistic AI-assisted coding assessments
• Guide developers through structured AI coding work inside the IDE
• Evaluate the quality of AI-assisted coding behavior
• Reduce blind acceptance of AI-generated code
• Capture reasoning, validation, testing, debugging, and review evidence
• Standardize AI coding practices across teams, hiring, training, and governance workflows
• Keep source code local by default while sharing only approved workflow evidence

Product modes

Savyre Code Intelligence includes three product modes.

Key features

• One extension for three Savyre product modes: Workflow, Evaluator, and Assessment
• 13-stage AI coding workflow from task input to final AI reflection
• Stage-specific prompts for Copilot, Cursor, Codex-style agents, Claude Code, and other AI assistants
• Developer ownership checklist to confirm understanding before submission
• Testing and validation evidence capture for commands, results, gaps, and failures
• Security and regression review prompts before final approval
• AI usage reflection to disclose meaningful AI assistance and limitations
• Protected-file guidance for secrets, credentials, and sensitive workspace files
• Local-first privacy posture where raw source code stays on the developer machine by default
• Assistant-agnostic workflow that can work across multiple coding assistants and technology stacks

Privacy and security

Savyre is designed for privacy-conscious AI-assisted coding workflows.

Code stays local by default

Your source code does not leave your local development environment in the default Savyre workflow. Savyre is designed to capture workflow evidence, summaries, validation status, and developer-approved outputs instead of uploading raw repositories.

What may be captured

Depending on your organization's configuration, Savyre may capture approved workflow evidence such as:

• Stage completion status
• Developer notes and summaries
• Files reviewed or changed, as metadata or file paths
• Test commands run and pass/fail status
• Validation gaps and known limitations
• AI usage reflection
• Final ownership statement

What is not submitted by default

By default, Savyre should not submit:

• Raw source code
• Full repository contents
• Secrets or credentials
• .env files
• Private keys, certificates, or tokens
• Unrelated workspace folders
• Protected files configured by the team

Configurable submission controls

Teams can configure what is included in submissions or reports. Raw code submission should remain disabled by default and should only be enabled through explicit organization policy and developer-visible configuration.

Recommended protected files:

.env
.env.*
*.pem
*.key
*.p12
*.pfx
secrets.*
credentials.*
node_modules/
dist/
build/

How it works

1. Install the Savyre extension.
2. Sign in or create a Savyre account.
3. Select a product mode: Workflow, Evaluator, or Assessment.
4. Open a coding project or assessment repository.
5. Start a Savyre session.
6. Follow the stage prompts inside the IDE.
7. Use your preferred AI coding assistant when needed.
8. Review and approve AI-generated suggestions before accepting them.
9. Run or document validation.
10. Complete the final summary, AI reflection, and ownership statement.
11. Submit only the approved evidence required by your product mode, team, or assessment.

Getting started

Common setup

1. Open a coding project in VS Code or your supported IDE.
2. Open the Savyre panel.
3. Sign in or create a Savyre account.
4. Confirm workspace and protected-file settings.
5. Select the product mode you want to use.

Savyre AI Coding Workflow

Use this mode when you want to guide day-to-day AI-assisted development through a structured process.

The workflow helps developers move through requirement analysis, codebase discovery, impact analysis, planning, implementation, testing, debugging, security review, final code review, and final ownership.

Typical flow

1. Select Savyre AI Coding Workflow.
2. Capture the task exactly as given.
3. Move through each Savyre workflow stage.
4. Use the current stage prompt with GitHub Copilot, Cursor, Codex-style agents, Claude Code, or another assistant.
5. Review every AI-generated suggestion before accepting changes.
6. Run relevant tests or document why tests could not be run.
7. Complete the final summary with changed files, validation evidence, known limitations, AI usage, and developer ownership.

Best for

• Engineering teams adopting AI-assisted development
• Developer productivity teams
• CTOs and engineering managers
• AI governance programs
• Individual developers who want a safer AI coding workflow

Savyre AI Coding Evaluator

Use this mode when you want Savyre to evaluate the quality of AI-assisted coding behavior, not just the final code.

The Evaluator combines the structured workflow with an evaluation layer. It can assess how the human developer uses AI across prompting, requirement understanding, codebase discovery, implementation, test design, debugging, code review, security thinking, and final ownership.

Typical flow

1. Select Savyre AI Coding Evaluator.
2. Start a workflow-backed evaluation session.
3. Complete the coding task through the Savyre stages.
4. Capture or approve evidence for prompts, files reviewed, tests run, review notes, debugging steps, and final reasoning.
5. Let Savyre evaluate AI-assisted coding signals such as:
   • Prompt quality
   • Requirement understanding
   • Codebase discovery discipline
   • Review of AI-generated code
   • Testing and validation depth
   • Debugging approach
   • Security and regression awareness
   • Over-reliance risk
   • Final explanation and ownership quality
6. Review the evaluator output for coaching, training, hiring review, or AI governance reporting.

Best for

• Engineering managers reviewing AI coding quality
• Learning and development teams
• Bootcamps and training programs
• Hiring teams that want deeper AI-coding signals
• Teams measuring responsible AI adoption

Savyre AI Coding Assessment

Use this mode when you want to run or evaluate a coding assessment where AI assistance may be allowed, controlled, or explicitly measured.

Savyre supports two assessment paths.

1. Savyre-created assessments

These are assessments created and managed by Savyre. The candidate opens the assigned repository or workspace and completes the task inside the IDE.

Typical flow

1. Select Savyre AI Coding Assessment.
2. Open the Savyre-provided assessment repository.
3. Confirm the assessment instruction, time limits, allowed tools, protected files, and submission rules.
4. Choose whether AI-assisted coding should be allowed, restricted, or evaluated as part of the assessment.
5. Complete the task through the Savyre workflow.
6. Submit approved evidence, validation results, final summary, AI usage reflection, and ownership statement.

2. Company-generated assessments

These are assessments created by the hiring company or customer. The candidate may receive a company-provided repository, task description, or coding challenge. Savyre helps structure the session, evaluate the work, and support proctoring and evidence capture.

Typical flow

1. Select Savyre AI Coding Assessment.
2. Open the company-provided repository or assessment workspace.
3. Load or confirm the company assessment instructions.
4. Configure the assessment mode, including whether Savyre should evaluate AI-assisted coding behavior.
5. Complete the task while Savyre tracks workflow progress, validation evidence, AI usage reflection, and final ownership.
6. Use Savyre outputs for technical evaluation, proctoring review, and assessment reporting.

Assessment options may include

• AI-assisted coding allowed or not allowed
• AI-assisted coding evaluated or not evaluated
• Candidate workflow evidence capture
• Proctoring support, depending on organization configuration
• Validation evidence and test results
• Final summary and ownership statement
• Evaluator signals for prompting, review, testing, debugging, and over-reliance risk

Best for

• Hiring teams and technical interviewers
• Recruiters and staffing agencies
• Startups and SMEs running practical coding assessments
• Enterprises evaluating AI-ready engineering talent
• Teams that want to assess both coding output and AI-assisted coding judgment

Workflow stages

Savyre uses a 13-stage workflow to keep AI-assisted coding structured and reviewable.

Supported AI coding assistants

Savyre is designed to work with common AI coding assistants and agent workflows.

Installation

From the VS Code Marketplace

1. Open Visual Studio Code.
2. Go to Extensions.
3. Search for Savyre Code Intelligence.
4. Click Install.
5. Open the Savyre panel from the activity bar or command palette.

From a VSIX file

1. Download the .vsix package.
2. Open VS Code.
3. Run Extensions: Install from VSIX... from the command palette.
4. Select the downloaded file.
5. Reload VS Code if prompted.

Local development

cd savyre-extension
npm install
npm run compile

Press F5 in VS Code/Cursor to launch the Extension Development Host. Set savyre.apiBaseUrl to http://localhost:5000 for local API dev.

Recommended repository structure

Savyre can create or use the following files in your workspace:

.savyre/
  README.md
  SAVYRE_INSTRUCTIONS.md
  config.json
  prompts/
    01-task-input.md
    ...
  tasks/
    01-task-input/README.md
    ...

AGENTS.md
.github/copilot-instructions.md
.cursor/rules/savyre-ai-workflow.mdc
SAVYRE_LOG.md

Commands

The exact command names may vary by extension version.

Select your product mode on the web at Settings → Savyre products or during extension onboarding.

Configuration

Savyre can be configured at the workspace level via .savyre/config.json and VS Code settings under Savyre.

Example workspace config shape:

{
  "configVersion": "1.0",
  "workflow": {
    "enabled": true,
    "mode": "workflow",
    "requireStageCompletion": true
  },
  "submission": {
    "includeRawCode": false,
    "includeFullRepository": false,
    "includeChangedFilesSummary": true,
    "includeValidationEvidence": true,
    "includeAiUsageReflection": true
  },
  "protectedFiles": [".env", ".env.*", "*.pem", "*.key"]
}

Developer operating rules

When working inside a Savyre session:

• Start with the task instruction, not with code.
• Use AI as a collaborator, not as the final authority.
• Review every generated change before accepting it.
• Keep changes scoped to the requirement.
• Prefer existing codebase patterns over new patterns.
• Run relevant tests or document why tests were not run.
• Investigate failures before asking AI to retry.
• Document assumptions, risks, validation, and open questions.
• Confirm final ownership before submission.

Completion standard

A Savyre stage is complete only when the developer can explain:

1. What was done
2. Why it was done
3. What files were reviewed or changed
4. What assumptions were made
5. What risks were considered
6. What tests or validation were performed
7. What remains open or uncertain

Final submission checklist

Before ending a Savyre session, confirm that:

• The task was captured accurately.
• Requirements and acceptance criteria were written down.
• Relevant codebase files and tests were discovered.
• Impact and risk areas were reviewed.
• The implementation plan was reviewed before coding.
• Code changes were scoped and reviewable.
• AI-generated code was reviewed and understood.
• Relevant tests or validation steps were run or explicitly documented as not run.
• Failures were investigated with a root-cause hypothesis.
• Security and regression risks were considered.
• The final diff was reviewed.
• The final summary includes changes, evidence, limitations, AI usage, and ownership.
• No raw source code, secrets, credentials, or protected files were submitted unless explicitly allowed by policy.

Known limitations

• AI assistant behavior may vary across Copilot, Cursor, Codex-style agents, Claude Code, and other tools.
• Some assistants may not automatically read Savyre prompts unless the prompt file is explicitly opened or referenced.
• Test execution depends on the project's local setup.
• Savyre cannot guarantee correctness of AI-generated code.
• Savyre does not replace human review, secure coding practices, or production approval processes.
• Product access may depend on your active Savyre subscription or organization entitlement.
• Savyre AI Coding Workflow team features are in preview; full team governance is rolling out incrementally.

Troubleshooting

The AI assistant is not following the Savyre workflow

Open or reference the current stage prompt explicitly. For Copilot, ensure .github/copilot-instructions.md is present. For Cursor, ensure the relevant rule file exists under .cursor/rules/.

The wrong product mode is selected

Open Settings → Savyre products on the web, or use the product cards on the Company tab in the extension panel.

Tests cannot be run locally

Document why they could not be run and add the closest available validation evidence.

You are unsure what data will be submitted

Review .savyre/config.json and the submission preview. Raw code should be disabled by default.

Release notes

0.2.1

• Evaluator: start evaluation in your open project folder (no clone)
• End session for Workflow and Evaluator with confirmation and toast
• Sign out clears local session state

0.2.0

• Rebranded extension to Savyre Code Intelligence
• One extension, three product modes: Workflow, Evaluator, Assessment
• Web products hub and extension product cards
• Workflow preview mode in the IDE

0.1.x

• Browser-based employer sign-in and device activation
• 13-stage workflow, metadata-only submission, proctoring
• Company assessment drafts and candidate invite flow

Support

For support, contact the Savyre team or visit savyre.com.

License

See LICENSE.