Skip to main content
Yagr logo
Yagr
Autonomous agent · deterministic workflows · orchestrator-ready

(Y)our(A)gent(G)rounded in(R)eality.

Most agents disappear into ephemeral scripts and blind API calls. Yagr takes the opposite path: it uses chat as the interface, but architects, validates, and deploys deterministic workflows underneath so the result stays auditable, inspectable, and durable.

Intent to workflowWorkflows as memoryBackend swappable
Start with YagrExplore n8n-as-code
Agent loop
$ npm install -g @yagr/agent@latest
$ yagr onboard
$ yagr start
$ yagr telegram onboarding
IntentStart from what to automate

Yagr should begin from the user goal, not from manually wiring raw nodes and scripts.

ReliabilityNo blind script sprawl

Yagr should not solve tasks by dropping opaque one-off scripts that nobody can audit tomorrow.

MemoryWorkflows are memory and muscle

Generated automations persist intent and execute it repeatedly as durable infrastructure.

Enginen8n is the current orchestrator

The orchestrator sits underneath the agent so the product layer can stay stable as runtimes evolve.

Quick Start

Get Yagr running before you read the manifesto.

If your first question is “how do I try it?”, the answer should be immediate: install the CLI, run onboarding once, then start the agent.

01Install

Install the published Yagr CLI globally with the package manager you already use.

02Onboard

Bind the current orchestrator, your model provider, and any optional integrations in one guided first-run flow.

03Start

Launch the agent and keep operating it through the same runtime loop from its own home.

npm install -g @yagr/agent
yagr onboard
yagr start

Install the CLI, run the onboarding once, then start the local agent runtime from its own home.

Open the getting started guideSee all CLI commands

Why Yagr exists

Magic chat for users. Reliable systems for engineers.

Yagr matters because autonomous behavior should not require opaque one-off scripts. The chat interface can feel magical, while the actual execution remains deterministic, inspectable, and built on strict workflow ontology.

🧭Intent first
Natural language becomes automation

Yagr is meant to turn user intent into real workflows, not just provide a nicer shell around existing commands.

🧩Yagrs compose
Use the node ecosystem as a tool palette

Yagr composes existing nodes into larger wholes instead of rebuilding integrations from generic HTTP glue every time.

🧠Workflows remember
Generated automations are durable memory

A workflow is persisted intent that Yagr can later inspect, explain, modify, and extend instead of starting from zero.

🏠Stable home
State lives outside arbitrary repos

Yagr uses a dedicated home directory so setup, linked surfaces, and runtime state do not leak into random working folders.

💬Gateways stay thin
Telegram and TUI are surfaces, not the brain

The agent logic lives above the gateways so Yagr can be reached through Telegram, local UI, CLI, or future web surfaces.

⚙️Orchestrator boundary
n8n today, other runtimes tomorrow

Yagr uses the n8n-as-code sync and schema foundation today while keeping the orchestrator replaceable for future runtimes.

How the repository is now organized

Yagr as the agent product, n8n-as-code as the engineering substrate.

For automation intent

Start from what you want to automate, not from raw node wiring. Yagr should be the product layer that translates that intent.

Read the Yagr starting point

For remote interaction

Use Telegram as one gateway into the same agent loop, with setup-managed credentials, onboarding links, and linked chats.

See the Telegram flow

For workflow engineering

n8n-as-code remains a product in its own right for workflow GitOps, AI skills, schema grounding, and TypeScript workflows.

Open the n8n-as-code product page

How the loop works

Intent, engine, workflow, memory.

Yagr is intentionally narrower than a generic assistant and intentionally higher-level than the workflow engineering stack. It sits in the middle: above the execution engine, below the user-facing intent, and connected to durable workflow artifacts it can evolve over time.

1

Express intent

Start with the automation you want, in natural language, instead of manually assembling raw implementation primitives.

2

Ground in the engine

Yagr uses the node and schema knowledge of the current backend so generation stays anchored to real capabilities.

3

Generate and validate

The agent produces workflows against the execution engine rather than embedding its own brain into the runtime.

4

Persist as memory

The resulting workflow becomes durable executable memory that Yagr can revisit, explain, and evolve later.

5

Operate through surfaces

TUI, Telegram, and future gateways remain thin surfaces over the same agent and engine boundary.

Start here

Start from the Yagr vision, then drop to the layer you need.

Use Yagr if you want the agent product that turns intent into automation. Use n8n-as-code if you want direct workflow GitOps, AI skill, VS Code extension, and TypeScript tooling.

Read the Yagr guideOpen n8n-as-code