Open source · MIT · 100% local
✦aura
Stop re-explaining yourself to every AI tool.
Define who you are — your stack, style, rules — once, in plain YAML files you own. aura serves that identity to Claude, ChatGPT, Cursor, and Gemini through the Model Context Protocol.
pip install aura-ctx click to copy
PyPI Downloads Stars MIT
The problem

Every AI tool starts from scratch. The industry is building solutions for this, but at the wrong layer.

Memory What happened in past conversations — Mem0, Zep, DeltaMemory not aura
Context engineering What the AI should know right now — LACP, Claudesidian, OpenClaw not aura
Identity Who you are, across everything — stack, style, rules, role ✦ aura

Memory is session history. Context is prompt engineering. Identity is who you are — structured, portable, and owned by you.

Quick Start
aura quickstart
$ pip install -U aura-ctx
$ aura quickstart

Step 1/5 — Scanning your machine...
  ✦ Detected 12 facts about your dev environment

Step 2/5 — Quick questions about you...
  What's your role?                   → Full-stack dev
  How do you want AI to talk to you? → Direct, no fluff
  Any rules or pet peeves?            → Always TypeScript strict
  ✦ Created developer (8 facts, 4 rules)

Step 3/5 — Configuring AI tools...
  ✦ Claude Desktop configured
  ✦ Cursor configured

Step 4/5 — Security audit...
  ✦ All clean — no secrets detected

Step 5/5 — Starting MCP server...
  ✦ http://localhost:3847/mcp
  Restart your AI tools — they know you now.

No Docker. No database. No cloud account. 30 seconds.

Before and after

Before
You: What do you know about me?

Claude: I don't have any information about you. Memory is turned off for your account.
After aura quickstart
You: What do you know about me?

Claude: Role: CS student · Editor: Cursor
Stack: TypeScript, Python, React, FastAPI
Rules: TypeScript strict — no 'any'
Supported tools
Claude Desktop
Streamable HTTP
auto
Claude Code
Streamable HTTP
auto
Cursor IDE
Streamable HTTP
auto
Windsurf IDE
Streamable HTTP
auto
VS Code
Copilot MCP
auto
Gemini CLI
SSE
auto
Codex CLI
Streamable HTTP
auto
ChatGPT
SSE / Dev Mode
manual

Any MCP-compatible client connects to localhost:3847.  What's MCP? →

Templates

14 built-in templates. Each includes typed facts and AI interaction rules tailored to the profile.

frontendReact, Next.js, Tailwind
backendFastAPI, SQL, Docker
data-scientistpandas, PyTorch, Jupyter
mobileReact Native, Swift
devopsK8s, Terraform, AWS
ai-builderLangChain, RAG, agents
founderMVP, speed, product
studentLearning, step-by-step
marketerCopy, SEO, growth
designerUX, Figma, accessibility
writerTone, style, audience
researcherMethodology, sources
$ aura templates
$ aura create mydev -t frontend
$ aura show mydev
Features
Auto-scan

Detects languages, frameworks, tools, editor, git identity. Incremental with SHA-256 hashing — only rescans what changed.

Context packs

Scoped YAML — developer, writer, work. Typed facts with confidence levels, freshness scoring, and TTL-based decay.

Smart delivery

3-level token delivery (~50/~500/~1000+). Priority scored by usage + freshness + confidence. Most sessions never exceed Level 1.

Per-agent permissions

Control which packs each AI tool sees. Cursor gets dev pack only. ChatGPT gets writer + work. Detected via HTTP headers.

Secret scanning

30+ credential patterns — AWS keys, GitHub tokens, API keys. Auto-redacts before serving. Last-resort scrub at serve time.

Import & export

Import from ChatGPT and Claude exports. Export to system-prompt, .cursorrules, CLAUDE.md, AGENTS.md, and more.

Pack health

Doctor checks bloat, stale facts, duplicates, secrets. Consolidation merges duplicates. Decay removes expired facts by TTL.

JSON Schema

Formal schema for context packs. Every pack validated on load. Inspect with aura schema or validate with aura validate.

Commands
Getting started
aura quickstartFull setup: scan → onboard → configure → audit → serve
aura scanAuto-detect your stack from tools, repos, and config files
aura onboard5 questions to generate your context packs
aura setupAuto-configure all detected AI tools
aura serveStart the MCP server on localhost:3847
aura serve --watchHot-reload on YAML changes
aura updateUpdate to the latest version
Managing packs
aura listList all context packs
aura show <pack>Display a pack's full contents with freshness scores
aura add <p> <k> <v>Add a fact without opening YAML
aura edit <pack>Open a pack in $EDITOR
aura diff <a> <b>Compare two packs side by side
Analytics & permissions
aura statsShow which facts your AI tools use most
aura stats --resetClear all usage counters
aura permissions set <agent> <packs>Restrict an agent to specific packs
aura permissions listShow current per-agent permissions
Health & schema
aura doctorPack health — bloat, stale facts, duplicates, secrets
aura auditScan for leaked API keys, tokens, credentials
aura audit --fixAuto-redact critical secrets
aura consolidateMerge duplicate facts, surface contradictions
aura decayRemove expired facts based on type-aware TTL
aura schemaPrint the JSON Schema spec for context packs
aura validate <pack>Validate a pack — field-level errors if invalid
Import & export
aura import -s chatgpt <file>Import from a ChatGPT data export
aura import -s claude <file>Import from a Claude data export
aura export <pack> -f system-promptUniversal LLM system prompt
aura export <pack> -f claude-mdCLAUDE.md section for Claude Code
aura export <pack> -f agents-mdAGENTS.md section for Codex, OpenClaw
Security
Localhost binding
Binds to 127.0.0.1 only — not reachable from the network.
Bearer token auth
Optional --token flag or AURA_TOKEN env var. SSE clients can pass via query param.
Scoped serving
--packs flag and per-agent permissions operate as two independent layers. Both must allow a pack for it to be served.
Read-only mode
AI tools read your context but can never write to it. --read-only.
Secret detection
30+ patterns — AWS keys, GitHub tokens, OpenAI/Anthropic API keys, database URLs, private keys. Auto-redaction at serve time.
No telemetry
No analytics. No crash reports. No usage tracking. Your YAML never leaves your machine unless you choose otherwise.
# Basic — localhost only
$ aura serve

# Locked down
$ aura serve --token s3cret --packs developer --read-only

9,200+ lines · 368 tests · 27 commands · 14 templates · MIT license

pip install aura-ctx && aura quickstart click to copy
GitHub PyPI Contribute Releases