Skip to content

SmartMur/smartmur-ai-core

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

72 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

SmartMur AI Core

SmartMur AI Core

Self-hosted AI operations platform that runs your infrastructure.
Your hardware. Your AI. Your rules.

CI Tests Python License Docker

Docs · Demo · Features · Install · CLI · Comparison · Contributing


Why this?

  • Eight subsystems, one platform. Skills, cron, messaging, SSH, browser automation, workflows, memory, and an encrypted vault -- integrated so a single YAML workflow can chain all of them.
  • AI-native, not AI-bolted. Built for Claude Code from day one. 42 MCP tools, an intake pipeline that decomposes requests into skill graphs, and a memory store that auto-injects context into prompts.
  • Zero cloud dependencies. Runs entirely on your hardware. No SaaS accounts required. Every credential stays in an age-encrypted local vault.

Quick Demo

$ claw status
Subsystem        Status    Details
───────────────  ────────  ──────────────────────────
Skill Registry   OK        14 skills loaded
Cron Engine      OK        3 jobs scheduled
Messaging        OK        Telegram, Slack connected
SSH Fabric       OK        4 hosts in pool
Browser Engine   OK        Chromium headless ready
Workflow Engine  OK        14 workflows available
Memory Store     OK        127 facts, 24 preferences
Vault            OK        age-encrypted, 8 secrets

$ claw skill run heartbeat
[heartbeat] Pinging 6 hosts...
  proxmox.local     OK   1.2ms
  truenas.local     OK   0.8ms
  docker01.local    OK   1.1ms
  pihole.local      OK   0.9ms
[heartbeat] Probing 3 HTTPS services...
  grafana.local     OK   cert valid 142 days
  homeassistant     OK   cert valid 89 days
  proxmox:8006      OK   cert valid 364 days
Exit code: 0

$ claw cron list
ID   Schedule       Type    Name                  Last Run
──   ──────────     ─────   ────────────────────  ──────────────
1    0 7 * * *      skill   morning-brief         2026-03-06 07:00
2    */15 * * * *   skill   heartbeat             2026-03-06 09:45
3    0 2 * * 0      shell   backup-verify         2026-03-03 02:00

$ claw ssh proxmox "qm list"
VMID  NAME             STATUS   MEM(MB)  BOOTDISK(GB)
100   docker01         running  16384    120
101   truenas          running  32768    32
102   k3s-master       running  8192     60

Features

Eight integrated subsystems, 42 MCP tools, 86 REST endpoints, 25 CLI command groups.

Subsystem What it does
Skill System 14 built-in skills. Registry, loader, auto-install from templates, SkillHub sync. Sandboxed execution with vault access control.
Cron Engine APScheduler + SQLite. Four job types: shell, claude, webhook, skill. Cron expressions, intervals, one-shot. Output routing to files or channels.
Messaging Gateway Telegram, Slack, Discord, email, iMessage. Notification profiles fan out to multiple channels. Inbound messages trigger skills.
SSH Fabric Paramiko connection pool with lazy creation and liveness probes. Multi-host execution, host groups, health checks. Home Assistant REST bridge.
Browser Automation Playwright + headless Chromium. Persistent session profiles. Navigate, screenshot, extract tables, fill forms, evaluate JS.
Workflow Engine YAML pipelines with five step types: shell, claude_prompt, skill, http, approval_gate. Conditions, rollback, notifications. 14 built-in workflows.
Memory Store SQLite-backed persistent memory. Categories: fact, preference, project_context, conversation_summary. Full-text search. 90-day auto-decay. Context auto-injection into Claude prompts.
Encrypted Vault age-encrypted credential store. Atomic writes. Rotation policies with warning/expired thresholds. Audit-logged. Sandboxed injection into skills.

Plus: a FastAPI dashboard (86 REST endpoints, 21 routers, HTMX frontend), an MCP server (42 native Claude Code tools), an intake pipeline (natural language to skill graphs), file watchers, and an append-only audit log.


The Compound Workflow

One YAML file. All eight subsystems. No other tool does this:

name: morning-infrastructure-check
steps:
  - name: check-vms
    type: shell
    command: "claw ssh proxmox 'qm list'"

  - name: diagnose-issues
    type: claude_prompt
    prompt: "Analyze this VM list. Flag any stopped VMs that should be running."
    input_from: check-vms

  - name: screenshot-dashboard
    type: shell
    command: "claw browse screenshot https://proxmox.local:8006 --output /tmp/pve.png"

  - name: log-finding
    type: shell
    command: "claw memory remember 'morning-check' '{{ steps.diagnose-issues.output }}'"

  - name: alert-operator
    type: skill
    skill: heartbeat
    notify: critical

  - name: approve-restart
    type: approval_gate
    channel: telegram
    message: "VMs need restart. Approve?"

  - name: restart-vms
    type: shell
    command: "claw ssh proxmox 'qm start {{ vm_id }}'"
    condition: "steps.approve-restart.approved"

SSH executes. Claude reasons. The browser screenshots. Memory stores. Messaging delivers. The approval gate waits for human judgment. The vault secures every credential. Cron runs it at 7 AM.


Architecture

                       You / Claude Code / Telegram Bot
                                  |
                       +----------+----------+
                       |    claw CLI (Click)  |
                       | claw-mcp (MCP/stdio) |
                       +----------+----------+
                                  |
     +-------+-------+----------+----------+-------+-------+-------+
     |       |       |          |          |       |       |       |
     v       v       v          v          v       v       v       v
  +------+ +-----+ +-----+ +------+ +-----+ +-----+ +-----+ +------+
  |Vault | |Skill| |Cron | | Msg  | | SSH | |Brow-| |Work-| |Memory|
  |(age) | |Reg. | |Eng. | |Gate- | |Fab- | |ser  | |flow | |Store |
  |      | |     | |     | | way  | | ric | |Eng. | |     | |      |
  +------+ +-----+ +-----+ +------+ +-----+ +-----+ +-----+ +------+
     |       |       |          |          |       |       |       |
     +-------+-------+----------+----------+-------+-------+-------+
                                  |
                       +----------+----------+
                       |  Config / .env       |
                       |  Audit log (JSONL)   |
                       |  File watchers       |
                       +---------------------+
                                  |
                  +---------------+---------------+
                  |               |               |
             Dashboard      Docker Compose   Telegram Bot
           (FastAPI+HTMX)   (5 services)    (AI responses)
            port 8200

Docker Compose Stack

Service Port Purpose
redis 6379 Pub/sub, cache, sessions
msg-gateway 8100 Multi-channel messaging API
dashboard 8200 Web UI + REST API
browser-engine 8300 Playwright automation API
telegram-bot -- Inbound message processing

Install

Quick start

git clone https://github.com/SmartMur/claude-superpowers.git
cd claude-superpowers
python3 -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"
cp .env.example .env   # add your tokens
claw status

Docker (full stack)

git clone https://github.com/SmartMur/claude-superpowers.git
cd claude-superpowers
cp .env.example .env   # add your tokens
docker compose up -d

Dashboard at http://localhost:8200. Browser engine at http://localhost:8300.


CLI Reference

claw provides 25 command groups:

Category Commands Description
Core skill, workflow, cron, memory, vault The eight subsystems
Infrastructure ssh, browse, msg, watcher Remote execution, browser, messaging, file monitoring
Orchestration intake, agent, orchestrate, dag, jobs Multi-agent dispatch, DAG execution, job management
Operations status, audit, report, benchmark, policy Monitoring, audit trail, reporting, safety policies
Management pack, template, setup, release, dashboard Bundles, config templates, setup wizard, releases
System daemon Cron daemon lifecycle (install/uninstall/status/logs)

Run claw <command> --help for subcommand details.


Comparison

vs. Bare Scripts + Cron

SmartMur AI Core Shell scripts
Credential management age-encrypted vault with rotation Plaintext in .env files
Scheduling APScheduler, 4 job types, output routing System crontab
Error handling Rollback actions, approval gates set -e and hope
Observability Audit log, dashboard, status command grep through logs
AI integration Native Claude Code + MCP None

vs. n8n

SmartMur AI Core n8n
Interface CLI + YAML + Claude Code Visual drag-and-drop
AI integration Claude-native (skills, MCP, prompts) AI nodes (bolt-on)
SSH Built-in pool with health checks Per-execution SSH node
Credential management age-encrypted vault with rotation Built-in credential store
Integrations 5 messaging channels + SSH + browser 400+ nodes
Target user Engineers who write code Business users who drag nodes

vs. Ansible

SmartMur AI Core Ansible
Scope Full operations platform (SSH + cron + messaging + AI + browser) Configuration management
AI reasoning Claude analyzes output, makes decisions None
Real-time response Inbound message triggers, file watchers Playbook runs only
Browser automation Playwright with session persistence None
Learning Memory store retains context across runs Stateless

vs. Agent Frameworks (LangChain, CrewAI)

SmartMur AI Core Agent frameworks
Type Ready-to-run platform Libraries for building agents
Time to first result pip install && claw skill run heartbeat Write agent code, configure tools, deploy
Infrastructure awareness SSH, Docker, Proxmox, Home Assistant Cloud-agnostic abstractions
Production runtime Cron, watchers, messaging, dashboard You build the runtime

Built-in Skills

Skill Purpose
heartbeat Ping hosts, probe HTTPS services, formatted status table
network-scan Network discovery and port scanning
ssl-cert-check TLS certificate expiry monitoring
docker-monitor Container health checks across Docker hosts
backup-status Backup job verification
container-watchdog Container restart loop detection and alerting
ops-report Operational status summary across all subsystems
deploy Deployment pipeline: git pull, pip install, docker build, health check
github-admin Repository audit, branch protection, security scanning
tunnel-setup Cloudflare tunnel token management
qa-guardian Code quality scanner: 12 checks across 4 categories
infra-fixer Docker infrastructure auto-remediation (40+ containers)
cloudflared-fixer Tunnel crash-loop detection and recovery
workspace-intake-smoke Intake pipeline smoke test

Create new skills: claw skill create --name my-tool --type python


Contributing

See CONTRIBUTING.md for development setup, coding standards, and PR process.

# Run the test suite
PYTHONPATH=. pytest --ignore=tests/test_telegram_concurrency.py

Security issues: see SECURITY.md.


License

MIT


Sovereignty is not a feature. It is the architecture.

About

SmartMur Core — AI-powered infrastructure automation platform with Claude Code integration, multi-agent orchestration, and self-healing capabilities

Topics

Resources

License

Code of conduct

Contributing

Security policy

Stars

Watchers

Forks

Packages

 
 
 

Contributors