strata-repo-intel 0.4.1

Local-first, budget-aware repository context compiler for safer AI-assisted coding.

Strata

Strata is a local repository-intelligence and safe AI coding workflow that builds focused context, traces related code, and helps review and apply AI-generated changes with clear scope controls.

Status: Strata is feature-complete for its current roadmap and has extensive automated validation. It is entering controlled real-repository UAT and release hardening; public production hardening is still in progress.

Why Strata

AI coding tools often see too little context, too much context, or the wrong context. Large repositories make that problem worse: important frontend, backend, test, workspace, and configuration relationships are easy to miss, while dumping the whole repo wastes tokens and can confuse the model.

Strata is built around a simple promise:

Understand the repository first. Give the AI only the most relevant context. Review and apply changes safely.

It combines repository intelligence, bounded context packs, guided workflow state, and patch review so developers can keep AI-assisted changes scoped, explainable, and intentional.

How It Works

1. Scan and understand the repository structure, files, symbols, imports, routes, and likely verification points.
2. Select and rank task-relevant files, symbols, relationships, and related tests.
3. Build a compact context pack for the configured AI workflow.
4. Review the proposed patch against expected scope, related files, and safety checks.
5. Apply intentionally, then verify through Strata's reports and your project checks.

Workspace and journey intelligence can connect related repositories and trace application flows across UI actions, API or message boundaries, backend handlers, services, and response paths. These are static explanations with confidence and gaps, not runtime traces.

Quick Start

The PyPI package is strata-repo-intel; the CLI command is strata. Strata requires Python 3.11 or newer; Python 3.11, 3.12, and 3.13 are supported, and Python 3.13 is the recommended development environment. The repositories it analyzes can target older runtimes.

Recommended install on Windows PowerShell:

pipx install strata-repo-intel
strata doctor install

From the repository you want Strata to understand:

cd path\to\your-repository
strata setup
strata start

strata start is the normal entry point. It shows current workflow status and one recommended next step. In an interactive terminal, it can continue step by step; repository-changing actions still require confirmation.

If you prefer the PowerShell bootstrap installer:

iwr https://raw.githubusercontent.com/Vaibhav-Malladi/Strata/main/install-strata.ps1 -OutFile install-strata.ps1
powershell -ExecutionPolicy Bypass -File .\install-strata.ps1

Installation Options

Install the published package for the current user:

python -m pip install --user strata-repo-intel

For local development from a cloned repository:

python -m pip install -e .

Windows users may also use the repository-local installer:

.\install.ps1

If Windows installs the strata command into the user scripts directory, reopen PowerShell or VS Code after installation so the updated PATH is detected. On older python versions, confirm that the Python user scripts directory is on PATH.

After installation, the normal workflow is:

strata start

For a small explicitly budgeted context pack:

strata context --budget small "your task"

For local development from a checkout, py is also supported:

py -m pip install -e .
strata doctor install
strata start

Recommended Workflow

The recommended workflow is the guided path:

strata start

To resume an interrupted or previously prepared session:

strata start --continue

Strata shows the current status and next recommended action during the guided workflow. Review the proposed scope and confirm any repository-changing action before Strata applies changes.

Settings

Use strata settings to review or change your workflow configuration. settings can be changed later, so initial setup choices are not permanent.

Examples:

strata settings set capability strong
strata settings set surface vscode
strata settings set mode hybrid

The guided workflow is the recommended path. advanced commands remain available for users who need direct control over context generation, review, apply, diagnostics, or verification.

Strata does not store API keys in the repository. It helps users configure keys in their own machine or user environment.

Generated Strata files live under .aidc/. Add .aidc/ to the analyzed repository's .gitignore; context packs, prompts, patches, and reports may contain code excerpts or task details.

Core Capabilities

Repository Intelligence

Strata discovers files, symbols, imports, routes, dependencies, framework hints, likely tests, and task-relevant starting points. It favors deterministic static analysis, bounded evidence, and explainable confidence over opaque indexing.

Focused AI Context

Context packs are budget-aware and task-ranked. Strata selects relevant source excerpts, symbols, related files, test hints, verification guidance, and relationship evidence while keeping Part I context artifacts as the token firewall. Representation tiers and caps help keep large repositories usable with both strong and smaller models.

Safe Patch Workflow

Strata is patch-first by default. It prepares context, collects or records an AI patch, reviews the patch, supports dry runs, blocks unsafe paths, warns about dirty trees or stale patches, controls allowed new files, and applies changes only through explicit confirmation. Application uses repository-relative paths, symlink escape protection, rollback on failure, and atomic file replacement where supported.

Frontend and Backend Understanding

Strata analyzes Python, JavaScript, TypeScript, Angular, React, and Go code. It can identify components, hooks, services, routes, API clients, backend route handlers, services, repository/database calls, response handling, and API boundaries with confidence-labeled evidence.

Workspace Intelligence

For applications split across repositories, Strata can represent related repositories, roles, explicit relationships, shared contracts, cross-repository references, and workspace dependency graphs. Workspace intelligence is designed to enrich context without silently inventing authoritative relationships.

User Journey Intelligence

Strata can explain user-facing actions as static journeys: UI entry points, frontend flow, API or message boundaries, backend handlers and services, data or external calls, response paths, confidence, diagnostics, and unresolved gaps. Journeys help an AI see the shape of a change without pretending to execute the application.

Supported Targets

Strata currently analyzes:

• Python
• JavaScript
• TypeScript
• Angular
• React
• Go

Strata itself is implemented primarily in Python. Java and Rust analysis are not currently supported. Some JavaScript, TypeScript, and framework inference is intentionally approximate, convention-based, and confidence-labeled.

Safety Model

Strata is designed to assist, not silently take control.

It runs locally by default, builds bounded selected context, keeps generated artifacts under .aidc/, and uses explicit review before apply. Patch validation checks for malformed diffs, absolute paths, traversal attempts, dangerous targets, symlink escapes, blocked new files, and scope mismatches. strata apply --dry-run validates without changing files; strata apply requires explicit intent and refuses to proceed on a dirty tree.

Strata does not commit or push. It does not guarantee a correct AI patch. Developers remain responsible for reviewing git diff, running project checks, and deciding what lands.

Secrets should remain in your machine or user environment. Strata stores environment variable names for API keys, not secret values in the repository configuration.

AI And Model Compatibility

Strata is adapter-neutral. It can support manual browser workflows, CLI-based assistants, IDE-oriented workflows, local models through Ollama, command adapters such as Codex CLI or Aider, and OpenAI-compatible HTTP endpoints.

The important design point is that users should not have to redescribe the model for every task. Strata uses configured adapter and workflow settings, then builds bounded context that remains useful across different model strengths and across model changes between sessions.

Current Maturity

Area	Status
Current roadmap implementation	Complete
Automated tests and safety gate	Strong
Synthetic multi-language coverage	Complete
Controlled real-repository UAT	Next
Public production release	Not yet

Strata should be treated as ready for controlled real-repository UAT, not as a fully production-proven autonomous coding system.

Limitations

Static analysis cannot resolve every runtime path, dynamic framework convention, dependency-injection edge, generated route, or reflection-heavy call. JavaScript, TypeScript, and framework extraction may be approximate. User journeys are static explanations, not runtime traces. Real-repository UAT is still pending.

Strata reduces context waste and review risk, but it does not eliminate hallucinations or guarantee that an AI-generated patch is correct. Review and validation remain part of the workflow.

Documentation

• Changelog
• Runtime compatibility
• V6 release notes
• V7 roadmap
• Frontend intelligence roadmap
• Backend intelligence foundation
• Workspace intelligence roadmap
• User journey intelligence roadmap

For CLI help, run:

strata help
strata help setup
strata help ask

Contributing

Strata does not currently have a separate contributing guide. For now:

• Open an issue before broad architectural changes.
• Keep changes focused and easy to review.
• Preserve safety checks and architecture invariants.
• Add focused tests for behavior changes.
• Avoid broad refactors mixed with feature or documentation work.

License

Strata is released under the MIT License.