Define agents in one file. Deploy them anywhere. Privacy-classified routing, multi-provider failover, formal verification — from prototype to production.
First product live: 6 AI agents on Telegram managing finances, health, calendar, groceries, notes, and daily briefings.
Real conversations with specialized agents on Telegram.
AgentiCraftBot
6 agents online
Live on Telegram — processing real messages through the full mesh stack.
Every piece of data is classified into one of four privacy levels — automatically, at the proxy layer, before it reaches any LLM provider.
Each agent has its own privacy rules, memory, and personality.
"spent 50 on coffee" — categorized, tracked, and summarized automatically.
L2Morning briefings, meeting prep, and schedule queries.
L1Shared shopping lists, inline editing, and family sync.
L0Medication tracking, reminders, and symptom logging.
L3Quick task capture with due dates, reminders, and completion tracking.
L0Daily synthesis: calendar + budget + tasks in one message.
L1This is the first product running on AgentiCraft. More agents across 12 domains are available via the craft CLI.
Add your own OpenAI, Anthropic, or any of 17 supported provider keys. Your queries go directly to the provider through our Rust proxy — we never see them.
Unlimited usage, free forever.
docker run -e TELEGRAM_BOT_TOKEN=xxx ghcr.io/agenticraft/bot
Run AgentiCraft on your own server. The full stack — 6 agents, privacy classification, Thompson Sampling routing — on a $9/month VPS.
From formal verification to user-facing products. Every layer has a clear boundary.
Fluent builder API with progressive complexity. Start simple, scale to advanced multi-agent orchestration.
agent = Craft.agent("analyst")
.model("gpt-5.4")
.capability("web_search")
.pattern("react")
.build()44 production services with defined SLAs. Security, deployment, agent marketplace, API gateway, and more built in.
mesh = Craft.mesh()
llm = await mesh.get_service("llm")
obs = await mesh.get_service("observability")
sec = await mesh.get_service("security")Define agents, topology, workflows, policies, and plugins in a single app.yaml. The runtime compiles it and deploys anywhere.
# app.yaml
agents:
- id: researcher
capabilities: [web_search]
models: { default: gpt-5-mini }
- id: analyst
capabilities: [data_analysis]
topology:
connections:
- { from: researcher, to: analyst }
policies:
guardrails: { pii: { action: mask } }The only platform that unifies AI agent logic with production infrastructure.
| AgentiCraft | LangGraph | CrewAI | Solo AgentGateway | Dapr Agents | |
|---|---|---|---|---|---|
| AI Agent Logic | |||||
| High-Performance Data Plane | |||||
| MCP + A2A Native | |||||
| Service Mesh Infrastructure | |||||
| Formal Verification | |||||
| Token Cost Controls | |||||
| Multi-Provider Failover | |||||
| L0–L3 PII Classification | |||||
| Consumer Product (Live) |
Most enterprises can build demos. Few can ship production agents.
One platform replaces your observability, reliability, cost, and security stack
44 production services — deploy, monitor, and scale with one command
Defense-in-depth security from day one
17 providers, open protocols (MCP, A2A)
Join the waitlist and get early access when we launch.
Launching soon.