Codasift watches your Python project locally, extracts routes, classes, and conventions via native AST in under 15ms, and keeps your .cursorrules, Claude XML, Copilot instructions, and Markdown context files automatically up to date.
class SessionService: def __init__(self, cache_ttl: int = 3600): self.default_ttl = cache_ttl # @router.post('/api/v1/session') [Protected]def create_session(self, user_id: str) -> Session: token = secrets.token_urlsafe(32) expiry = int(time.time()) + ttl return Session(token, user_id, expiry)- def create_session(user_id) [Protected] - Coding style: snake_case, strict type hints
You save a Python file in your workspace. Watchdog coalesces the event locally in 1.0s.
.cursorrules Go StaleAI coding assistants are only as good as the context you feed them. When your architecture changes, stale instructions cause phantom methods, wrong route guards, and broken imports.
[Protected] route dependencies.ASTNormalizer strips formatting to update AST in RAM with zero AI API calls.@router.post endpoints as Public vs Protected automatically.Uploading proprietary codebases, database models, and internal routes to third-party indexing servers creates security vulnerabilities. Codasift parses everything locally on your CPU via native Python AST (cloud collaboration is strictly opt-in).
Up to 80% of codebase saves are cosmetic: reformatting, adding docstrings, or tweaking comments. Codasift's dual-gate classifier updates the structural map in RAM and skips the AI model entirely.
You might brainstorm in Claude, edit in Cursor, and autocomplete with Copilot. Instead of maintaining three conflicting snapshots, Codasift updates all 4 context files atomically on file save.
Codasift does not require proprietary editor extensions. It extracts your architecture via native Python AST and atomically writes the exact context files your AI tools already read.
# .cursorrules - Auto-Generated by Codasift (on-device AST)
# Read automatically by Cursor IDE on every prompt
## How to Run & Test
- Run tests: pytest tests/ -v
- Start API: uvicorn main:app --reload --port 8000
## Tech Stack & Conventions
- Python 3.13, FastAPI, Pydantic v2, SQLAlchemy 2.0
- Coding Style: snake_case for functions/vars, PascalCase for classes
- Quotes: double quotes ("...") | Type annotations: required on all signatures
## API Endpoints & Access Control
- POST /api/v1/sessions [Protected] -> SessionService.create_session(user_id: str, ttl: int = 3600)
- POST /api/v1/checkout [Protected] -> initiate_checkout(session: Session, items: list[Item])
- GET /health [Public] -> health_check()
## Recent Semantic Changes
- auth/session.py: Added configurable ttl parameter (default 3600s) to create_session.* Empirical benchmarks measured in standardized CPython 3.13 test suites across Python codebases up to 15,000 lines (see Terms of Service §01.E & §07.E). Actual execution speed is hardware-dependent. All benchmarks execute 100% locally on-device with zero telemetry.
Keep Cursor, Claude, Copilot, and custom agents synchronized with your live Python codebase on save. 100% on-device by default with zero code upload, and instant.