Your AI. Your workflow.

Meet your new personal AI agent.

A practical assistant for coding, research, and the work in between. OpenArva brings capable AI models, useful tools, and thoughtful safety controls into one personal workspace.

Start with OpenArva Explore on GitHub
Local or cloud models You stay in control Open source
openarva — personal workspace
you ❯ Help me understand this project --domain coding openarva › Looking through the project... ✓ Checking workspace files ✓ Finding key modules ✓ Gathering relevant context I found a TypeScript application with a CLI entry point, an agent runtime, and a set of workspace tools. Here’s how the pieces fit together... Task complete · workspace stays yours
Safety, built inApprovals for consequential actions
Memory on your deviceLocal SQLite storage and retrieval
Your model, your choiceConnect local or cloud providers
Open sourceInspect it, run it, make it yours

Made to be useful

One agent. Your kind of work.

Bring your own models and tools. OpenArva helps turn a request into a plan, carries out approved steps, and gives you a clear result.

Work with your code

Search project files, understand a codebase, and run allowlisted commands in a bounded workspace environment.

Research with context

Retrieve local notes and task history, fetch public web pages with network safeguards, and explore public GitHub activity.

Keep useful memory

Store task outcomes, preferences, and indexed documents in a local SQLite memory store you can manage.

Choose the model

Route tasks to supported cloud providers or connect to local Ollama and LM Studio models.

Resume when ready

Persist autonomous task progress, verify results, and explicitly resume recoverable work when needed.

Stay in control

Tool inputs are validated, permissions are checked, and higher-risk actions require human approval.

Bounded autonomy

From “can you help?” to work you can verify.

OpenArva structures autonomous tasks as a controlled workflow—not a black box. It plans with available tools, asks before consequential actions, and keeps a record so you can inspect what happened.

Explore the architecture
01

Understand

Take in the goal and retrieve relevant local context.

02

Plan

Break the request into steps using tools the agent actually has.

03

Act with permission

Validate tool inputs and request approval when the action requires it.

04

Verify and recover

Check results, record progress, and replan or resume when appropriate.

Connect what you use

Real integrations, clearly labeled.

OpenArva includes working connectors and a wider catalog for future adapters. A catalog entry does not mean an integration is already available.

Available connectors

Implemented
Telegram WhatsApp via Twilio GitHub crawler Public web research

Each connector has its own setup and access requirements. See the capability matrix for details.

Connector catalog

Scaffolds included
Slack Microsoft Teams Matrix LINE Mastodon Notion Linear + more in catalog

Many listed platforms are interface placeholders, not live integrations. We label them honestly rather than implying they work.

Your setup, your boundaries

Personal AI should respect your choices.

Run OpenArva with a local model, use a configured cloud provider, or choose local-only mode. Memory is stored locally by default; you decide which model receives each request.

Read the security guidance
Choose local-onlyReject cloud model routing when you enable local-only mode.
Keep memory localTask context and vector records live in your local SQLite store.
Approve consequential actionsPermissions and approval gates help keep you in the loop.
Bring credentials safelyUse environment variables or an ignored local .env file; never commit secrets.

Up and running in minutes

Make OpenArva yours.

Install the CLI, launch the setup wizard, and pick a provider. To use cloud models, set the matching API key in your environment. For local inference, connect Ollama or LM Studio.

# Install OpenArva
npm install -g openarva

# Configure your agent
openarva init

# Start with a task
openarva run --domain coding --instruction "Explain this project"

Good to know

Questions, answered.

What is OpenArva?

OpenArva is a personal AI agent you can run from your computer. It brings supported AI providers, local memory, workspace tools, and optional messaging connectors together behind permission and approval controls.

Do I have to use a cloud AI provider?

No. OpenArva supports local model endpoints such as Ollama and LM Studio. Enable local-only mode to block cloud routing; local inference still requires an available local model that can perform the requested task.

Does OpenArva connect to every platform in its catalog?

No. The catalog includes many scaffolds and interface placeholders. Working integrations are identified separately in the project capability matrix; a catalog listing alone does not mean a service adapter is implemented.

Where does OpenArva store memory?

Local vector memory is stored in SQLite in your home directory under ~/.openarva/memory/. You can index local documents and clear the stored memory with the CLI.

Is OpenArva fully autonomous?

OpenArva can plan and execute bounded tasks using registered tools, but it is not AGI or ASI. Tool permissions, approvals, timeouts, and verification limit what it can do, and you remain responsible for reviewing consequential changes.

What do I need to get started?

Use a supported Node.js version and npm, install the CLI, and run openarva init. Configure either a supported provider credential or a reachable local model endpoint. See the README for detailed setup guidance.

Put a personal agent to work.

Start small, choose your model, and keep the controls in your hands.

Get started View source