The AI That Scales With You.
One Mind. All Channels. Continuous Relationship.
open-source
self-Hosted
security-first
Discord
Telegram
Slack
WhatsApp
Signal
Teams
Web
Guards
Echo Guard Content Guard Skill Guard Agent Guard
ScalyClaw ScalyClaw Node
Orchestrator Memory Models Agents Skills Vault MCP Scheduler
Queue
Worker 1
Code · Skills · Commands
Worker 2
Code · Skills · Commands
Worker 3
Code · Skills · Commands
+ N
Dashboard

Real-time system health at a glance — node status, models, budget, channels, and workers, all auto-refreshing.

ScalyClaw Dashboard
Overview
Mind
Usage
Channels
Models
Agents
Skills
Memory
Vault
MCP
Scheduler
Engagement
Security
Logs
Workers
Jobs
Overview
Connected

Overview

↻ Refresh
Node
Hostnamemac-studio
Addresslocalhost:3000
Uptime6h 32m
Version1.4.2
Memory128 MB
Models
Chat3
Embedding1
Usage
Messages312
Input Tokens271,300
Output Tokens16,700
Total Calls594
Budget
Month Cost$12.4832
Limit$50.00
ModeHard
of $50.00
Agents
Count3
Skills
Count5
Memory
Stored847
Channels
telegramdiscordslack
MCP
Servers2
Tools17
Workers
Count2
Scheduler
Active Jobs4
Jobs
Active12

Mind

Save
IDENTITY.mdSOUL.mdUSER.md

Usage

⚙ Settings ↻
Today7d30dThis MonthAll Time From To
Today's Cost
$2.48
of $5.00/day
Month Cost
$12.48
of $50.00/mo
Total Cost
$12.48
Input Tokens
271,300
Output Tokens
16,700
API Calls
594
By ModelBy Type
ModelProviderIn TokensOut TokensIn CostOut CostTotalCalls
claude-sonnet-4-6Anthropic98,40034,300$0.30$0.20$0.50312
gpt-4oOpenAI71,20023,300$0.18$0.10$0.28187
gemini-2.0-flashGoogle38,7005,400$0.02$0.01$0.0395
Daily Cost
Daily Tokens

Channels

Save
TelegramDiscordSlackWhatsAppSignalTeams
Enabled
Bot Token
Allowlist (User IDs)
DM Policy
Group Policy
Text Chunk Limit
Media Max MB

Models

+ Add Model + Add Embedding ↻
Chat ModelsEmbedding Models
IDNameProviderTempMax TokContextPriWtIn $/MOut $/MCapabilitiesOn
cs46claude-sonnet-4-6Anthropic0.74096200K11$3$15tool image✎ 🗑 Test
gpt4ogpt-4oOpenAI0.74096128K21$2.5$10tool image✎ 🗑 Test
gf20gemini-2.0-flashGoogle0.581921M31$0.1$0.4tool✎ 🗑 Test

Agents

+ New Agent ↻
IDNameDescriptionModelsSkillsToolsMCPsOn
researchresearch-agentWeb research & summarizeautoweb-search8filesystem✎ 🗑
codecode-agentWrite & review codeclaude-sonnetexec-js exec-py12—✎ 🗑
skill-creatorskill-creatorCreates new skillsauto—5—locked

Skills

⬆ Upload ↻
IDNameDescriptionLangOn
web-searchWeb SearchSearch using Brave APIjs📄 🗑
summarizeSummarizerCondenses text to bulletspython📄 🗑
exec-codeCode RunnerExecute arbitrary codejs📄 🗑

Memory

+ Store ⚙
SubjectTypeTagsConfidenceUpdated
▸User prefers dark mode UIfactpreferenceHighFeb 24🗑
▸Project uses Bun runtimefactprojectHighFeb 23🗑
▸Alice's deadline is March 1analysispersonLowFeb 22🗑

Vault

+ Add Secret ↻
NameValueActions
OPENAI_API_KEY••••••••Reveal Copy Delete
ANTHROPIC_API_KEY••••••••Reveal Copy Delete
TELEGRAM_BOT_TOKEN••••••••Reveal Copy Delete

MCP Servers

+ Add Server Import JSON ↻
IDTransportStatusToolsOn
▸filesystemstdioconnected12✎ ↻ 🗑
▸web-searchhttperror0✎ ↻ 🗑
▸code-runnerssedisconnected5✎ ↻ 🗑

Scheduler

+ Reminder + Recurrent ↻
RemindersTasks
ActiveAllCompletedCancelledFailed
IDTypeDescriptionCronNext RunStateCreated
r-001one-timeCheck in with team—Feb 24 14:00ActiveFeb 24✓ 🗑
r-002recurrentDaily standup reminder0 9 * * *Feb 25 09:00ActiveFeb 20✓ 🗑
r-003one-timePay rent——CancelledFeb 18🗑

Engagement

Trigger Now Discard Save
GeneralQuiet HoursTriggers
Enabled
Model
Cron Pattern
Idle Threshold (min)
Cooldown (sec)
Max Per Day

Security

Discard Save
Message GuardSkill GuardAgent Guard
Enabled
Model
Layer 1 — Echo Guard
Similarity Threshold
Layer 2 — Content Guard
Checks for injection, social engineering, and harmful content

Logs

↻
NodeWorkers
1m5m10m30m1hRange
allinfowarnerrordebug
10:32:14.521 INFO [orchestrator] message received channelId=telegram:12345
10:32:14.892 INFO [tool] exec_code started jobId=98
10:32:15.103 ERROR [worker] connection timeout after 5000ms
10:32:15.440 WARN [guard] echo similarity 0.91 — blocked repeated message
10:32:16.001 INFO [orchestrator] tool result received jobId=98
10:32:16.312 INFO [channel] sending reply channelId=telegram:12345 chunks=2
10:32:17.891 INFO [memory] stored 3 new facts from conversation
10:32:18.440 WARN [budget] daily spend $2.48 approaching threshold (80%)

Workers

↻ Refresh
mac-studio.local
localhost:4001
StartedFeb 24 08:00
Uptime6h 32m
Concurrency4
Version1.4.2
web-server-01
192.168.1.10:4001
StartedFeb 24 07:30
Uptime7h 02m
Concurrency8
Version1.4.2

Jobs

⚙ ↻ Auto-refresh
AllMessagesAgentsToolsProactiveSchedulerSystem
activewaitingcompletedfailed (2)delayed
IDNameQueueCreatedProcessed
48291process-messagemessages14:01:2214:01:23👁 ✓ ✗
48290exec-codetools14:01:1014:01:11👁 ✓ ✗
48285delegate-agentagents13:58:4413:58:45👁 ✓ ✗
Pending Messages
ChannelMessagesTypesOldest
telegram:123452text14:01:20
1 / 16
Features

Persistent Memory

Hybrid vector + full-text search. Your AI automatically extracts and recalls facts from every conversation. It truly learns you.

Every Channel

Discord, Telegram, Slack, WhatsApp, Signal, Teams, or the built-in gateway. One assistant, one memory, all your conversations.

Autonomous Agents

Delegate complex work to specialized sub-agents with their own models, tools, and skill access. They think so you don't have to.

Extensible Skills

Write skills in JavaScript, Python, Rust, or Bash. Auto-dependency install, hot-reload, and zip deployment. Your AI can even write its own.

Triple-Layer Security

LLM-powered guards validate every message, skill, and agent. Echo guard, content guard, and code validation. Fail-closed by default.

MCP Integration

Connect any Model Context Protocol server. Tools are discovered automatically and become first-class citizens. Zero glue code required.

Scalable Workers

Deploy workers anywhere. Same machine or remote, they auto-connect via Redis and BullMQ. Scale execution without touching the core.

Zero-Downtime Reload

Skills, agents, config, and MCP servers reload live via Redis pub/sub. Update anything without restarting. No interruptions, ever.

Budget Control

Monthly and daily limits with hard/soft enforcement. Per-model tracking, configurable alert thresholds. Never get a surprise bill again.

Security

Security by Default

Three LLM-powered guards, an encrypted vault, and container-ready workers. Every layer is designed to fail closed — if anything looks wrong, it gets blocked.

  • Echo Guard detects prompt injection attacks
  • Content Guard blocks harmful requests
  • Skill & Agent Guards validate code and definitions
  • Vault stores secrets in Redis — never exposed in messages, memory, or logs
  • You can run Workers in containers for full execution sandboxing
1

Echo Guard

LLM repeats the message; if similarity < 0.9, injection detected. Blocked.

2

Content Guard

Analyzes for social engineering, harmful content, jailbreak attempts.

3

Skill & Agent Guard

Validates code for malicious payloads and excessive permission requests.

4

Vault

Secrets stored in Redis, injected as env vars at runtime. Never echoed in messages or memory.

5

Container Sandboxing

Run workers in Docker containers. Code execution is fully isolated from the host.

Built With ♥

BBun
TypeScript
Redis
SQLite
React
Fastify
BullMQ
Tailwind