openAgent
gitlabThe open source UNIVERSAL agent.
https://gitlab.com/bitzzdev/openAgent.git- TypeScript61.1%
- MDX21%
- TSX14.5%
- CSS2.9%
- HTML0.1%
Overview
OpenAgent is a universal, local-first AI agent platform designed to run autonomously on your machine. Beyond software development, OpenAgent is capable of reasoning about arbitrary workflows—integrating with APIs, cloud services, databases, research pipelines, and local environments through a dynamic, persistent tool architecture.
OpenAgent features a shared core runtime accessible via:
- Interactive Terminal TUI (
openagent) - Local Web UI (
openagent web) - Headless & Scripting CLI (
openagent run "...")
Architecture
OpenAgent
│
┌───────┴────────┐
│ Agent Runtime │
└───────┬────────┘
│
┌────────────────────┼────────────────────┐
│ │ │
▼ ▼ ▼
TUI Web UI CLI
│ │ │
└────────────────────┼────────────────────┘
│
▼
Agent Engine
│
┌─────────────────┼─────────────────┐
│ │ │
▼ ▼ ▼
Tool Registry Credential Broker Permission System
│
┌────┴────┐
▼ ▼
Built-in Generated
Tools Tools
│
▼
Sandbox
Key Features
1. Dynamic Tool Creation & Persistent Registry
When OpenAgent encounters a task requiring an external capability it does not possess, it autonomously creates, validates, sandbox-tests, and registers a persistent tool.
- Built-in tools:
read,write,edit,glob,grep,bash,webfetch,websearch,question,todowrite,skill,tool_generate. - Agent-generated tools: Dynamic tools saved across sessions in
~/.openagent/tools/. - Tool Management: Inspect, test, enable, or remove tools via
openagent toolsor/tools.
2. Credential Broker & Strict Secret Isolation
- Zero Secret Leakage: The LLM model context never receives the raw credential vault.
- Scoped Injection: Generated tools only receive credentials they explicitly declare and have been authorized for.
- Automatic Redaction: API keys, tokens, and passwords are automatically masked with
[REDACTED]across logs, session history, TUI/Web output, and error traces. - Management: View and configure credentials via
openagent credentialsor/credentials(without exposing secret values).
3. Multi-Provider AI Support
Connect to leading AI providers or run offline:
- Anthropic (Claude 3.7 Sonnet / Opus / Haiku)
- OpenAI (GPT-4o, o1, o3-mini)
- Google Gemini (Gemini 2.0 Flash / Pro)
- Groq, Mistral, xAI, Amazon Bedrock, OpenRouter, Ollama & local models.
4. Fine-Grained Permission System
Control agent autonomy with clear permission tiers (allow, ask, deny):
filesystem.read/filesystem.writeshell.execute/process.spawnnetwork.requestcredential.readtool.generate
Quickstart
Running the Terminal TUI
openagent
Running Local Web Mode
openagent web
Starts the local OpenAgent server at http://localhost:4096 and opens the browser UI.
Running Headless Tasks
openagent run "Summarize recent Git commits and check for any outdated dependencies"
CLI Usage
# General help and commands
openagent --help
# Dynamic Tool Management
openagent tools list # List built-in and generated tools
openagent tools inspect <tool_name> # View tool schema, source, and credentials
openagent tools remove <tool_name> # Delete a generated tool
openagent tools enable <tool_name> # Enable a tool
openagent tools disable <tool_name> # Disable a tool
# Credential Management
openagent credentials list # List configured keys and dependencies
openagent credentials set <KEY> <VALUE> # Securely store a credential
openagent credentials remove <KEY> # Remove a credential
# Models and Providers
openagent models # List available models
openagent providers # Configure AI providers
Development & Testing
# Install dependencies
bun install
# Run the OpenAgent dev CLI
bun dev
# Run unit and security tests
bun test
Upstream Attribution
OpenAgent builds upon and extends the open-source foundation originally created by OpenCode. We are grateful to the original creators and maintainers.
License
MIT License. See LICENSE for details.