agentkit-runtime 0.1.2

Agent Runtime SDK for igentbase — build sandboxed AI agents with one import

agentkit-runtime — Agent Runtime SDK for igentbase

Build sandboxed AI agents with 38 permissions, encrypted vault, scoped storage, and sandboxing.

Install

pip install agentkit-runtime

Usage

from agentkit_runtime import AgentRuntime

agent = AgentRuntime()

# Network requests (declared hosts only)
resp = agent.internet.get("https://api.openai.com/v1/models")

# Encrypted secrets vault
api_key = agent.secrets.require("OPENAI_API_KEY")

# Clipboard
text = agent.clipboard.read()

# Scoped storage (declared directories only)
files = agent.storage_read.list_files("Documents", "*.pdf")
content = agent.storage_read.read_text(files[0].path)

# Notifications
agent.notifications.send("Done!", "Processed all files")

# Logging (local file + stderr + syslog)
agent.local.log.info("task complete")

# Push log to Manager UI
agent.system.log.info("Agent ready")

# Check permissions at runtime
if agent.permissions.is_granted("CAMERA"):
    photo = agent.camera.capture()

# Check agent dependencies
if agent.dependencies.is_installed("com.dev.other-agent"):
    result = agent.call("com.dev.other-agent", "analyze", {"text": "hello"})

Permissions

Agents declare permissions in manifest.yaml. Users review and approve at install time. Runtime permissions prompt the user on first use (Allow / Allow Once / Don't Allow).

38 permissions across 14 categories:

Category	Permissions
Network	INTERNET, LISTEN_PORT
Display	CLIPBOARD, SCREEN_CAPTURE
Device	CAMERA, MICROPHONE, AUDIO_OUTPUT
System	NOTIFICATIONS, SYSTEM_INFO, ENV_VARS, SECRETS, RUN_COMMAND, OPEN_URL
Inspection	NETWORK_INFO, INSTALLED_SOFTWARE, PROCESS_LIST, SERVICE_LIST
Modification	INSTALL_SOFTWARE
Storage	STORAGE_READ, STORAGE_WRITE, FULL_STORAGE_READ, FULL_STORAGE_WRITE, AGENT_SHARED_DATA
Hardware	BLUETOOTH, LOCATION, USB
PIM	CONTACTS, CALENDAR
Power	WAKE_LOCK, SYSTEM_SETTINGS, WALLPAPER, VIBRATE
Connectivity	NFC
Security	BIOMETRIC, ACCESSIBILITY
Scheduling	SCHEDULED_TASKS
Service	BACKGROUND_SERVICE, RECEIVE_EVENTS

Security

Agents run in a secure, isolated environment. Every system resource is access-controlled through permissions — agents can only do what the user explicitly allows.

Docs

• Developer Guide: https://www.igentbase.com/docs-developer
• Provider Reference: https://www.igentbase.com/docs-providers
• Marketplace: https://www.igentbase.com

---

Built by igentbase Inc.