Why it matters
Nosis provides developers with a flexible framework for building local personal agents, emphasizing controlled execution and extensibility. Its local-first approach and plain file persistence offer a unique alternative for managing agent data and operations.

What changed

EvannZhongg has released Nosis, an open-source personal agent runtime designed for local-first operation. The project, available on GitHub, provides a unified agent core accessible through multiple interfaces: a terminal user interface (TUI), a browser-based graphical user interface (GUI), and integration with Feishu for messaging.

Key features of Nosis include:

  • Controlled Local Execution: Sessions can be configured with permission presets like Ask for approval, Workspace Access, or Full Access. Shell commands are sandboxed, and file operations are confined to the current workspace.
  • Desktop Automation: The agent can interact with the desktop via computer_screenshot and computer_action tools, enabling actions like clicks, key presses, and scrolling, with user approval based on the current permission preset.
  • Web Research Capabilities: Integrated tools like web_fetch and web_search allow agents to access public web pages and perform searches using providers such as DuckSeek, Exa, or DeepSeek.
  • Agent Loop Features: Nosis supports streaming responses, tool batching, context compaction, steer/cancel commands, plans, and background jobs.
  • Resumable Sessions: Conversation and execution events are logged to an append-only JSONL journal, allowing sessions to be resumed across different frontends.
  • Memory and Scheduling: The runtime includes global and workspace-scoped long-term memory, alongside support for one-shot, interval, and cron-based scheduled tasks.
  • Model and Media Handling: It utilizes OpenAI-compatible endpoints via the native OpenAI SDK and LiteLLM for other routes and image generation, allowing flexible provider routing for different agent roles.
  • Extensibility: Nosis supports standalone skills, plugins that provide agent capabilities, and standalone MCP servers.
  • Configuration and Data Persistence: All configurations, sessions, memory, and artifacts are persisted as plain files, primarily within the ~/.nosis/ directory, facilitating local management and version control.

The project is built using Python and Node.js, with requirements including Python >= 3.11 and Node.js >= 22. Installation involves cloning the repository, installing Node.js dependencies, building the frontend assets, and installing the Python server extras.

Why it matters for builders

Nosis offers builders a robust and flexible local runtime for developing personal AI agents. The emphasis on local-first operation and plain file persistence provides developers with greater control over data and execution environments, which is crucial for privacy-sensitive applications or offline agent functionality. The layered architecture, separating the agent core from frontends and the runtime, promotes modularity and easier integration of custom tools and interfaces.

Practical impact

Developers can leverage Nosis to build sophisticated personal agents that can automate desktop tasks, perform web research, and manage complex workflows locally. The clear separation of concerns and the provided SDKs (@nosis/client TypeScript SDK and a public stdio protocol) enable third-party frontends to integrate with the Nosis agent core. Builders can explore the detailed configuration options for providers, tools, and agent behavior stored in files like provider_config.json and agent_config.json to customize agent capabilities. The quick start guide provides clear steps for setting up the TUI and server components, allowing for immediate experimentation with the agent's functionalities.

Caveats and source limits

This release is described as a "fresh release" with version v0.1.1, indicating it is an early-stage project. Specific performance benchmarks, detailed comparisons against other agent runtimes, or extensive user adoption metrics are not provided in the source material. The source focuses on the technical architecture and capabilities of the Nosis runtime itself, rather than independent validation or broad market impact. Information regarding pricing or commercial support is also absent, as it is an open-source project.

Sources

Written with AI assistance from the linked sources; every claim below was checked against them automatically. How we produce articles.

Claim check: 6/6 supported claims - 6 evidence links - 100% avg confidence
  • Nosis is an open-source, local-first personal agent runtime.supported - github.com
  • Nosis offers a TUI, browser GUI, and Feishu messaging access.supported - github.com
  • Nosis supports per-session permission presets, OS-sandboxed shell execution, and desktop control.supported - github.com
  • Nosis persists sessions, plans, and memory as plain files.supported - github.com
  • Nosis requires Python >= 3.11 and Node.js >= 22.supported - github.com
  • The latest release of Nosis is v0.1.1.supported - github.com

Caveats

  • Single-source caution: verify critical details at the linked source.
Radar score 87/100 - how it was calculated
Reliability82
Freshness92
Novelty77
Technical85
Developer96
Ecosystem72
Confidence100
  • Reliability 82: GitHub metadata supports source trust
  • Freshness 92: Fresh GitHub activity
  • Novelty 77: Fresh GitHub release
  • Technical 85: Repository technical metadata
  • Developer 96: Developer tooling signals
  • Ecosystem 72: Fresh GitHub release
  • Confidence 100: Claims have reliable evidence
Share
XLinkedInHacker News

Discussion

Loading comments...

Related articles

Agents - Nov 22, 2026AgentForge: Open-Source AI Coding Agent HarnessAgentForge is an open-source terminal-based harness for AI coding agents, designed for studying agent loops, tools, and persistence. It supports multiple AI models and provides a framework for developing and testing agent behaviors.Agents - Nov 20, 2026Grajo: Self-Hosted Agentic Framework and MCP PlatformGrajo is an open-source, self-hostable agentic framework and Model Context Protocol (MCP) platform. It allows users to build and run AI agents with persistent memory, research capabilities, and P2P sharing, while also serving as an MCP platform to connect existing AI clients like Claude Code and Cursor.Agents - Nov 22, 2026Waku Agent: Open-Source AI Agent Harness with Local-First ArchitectureWaku Agent is an open-source AI agent harness designed for local-first operation, featuring a readable codebase that emphasizes core agent pillars: harness, loop, memory, eval/LLM-Ops, and an always-on layer. It offers local data storage, a straightforward Python loop, and integrated evaluation tools.Agents - Nov 20, 2026Taskuary: Open-Source AI Work Partner for Inbox ManagementTaskuary is a new open-source AI work partner designed to manage emails, chats, and tickets by consolidating them into a single, ordered list. It employs AI agents to handle tasks like research and drafting, with a human-in-the-loop approval step before any outgoing action.Agents - Nov 22, 2026CompozyOS: AI Agent Operating SystemCompozyOS is an operating system designed for AI agents, acting as an autonomous control plane. It allows users to see and steer sessions from tools like Claude Code, Codex, and Cursor, with agents checking in autonomously and users intervening only when necessary.Agents - Nov 20, 2026PraisonAI v4.7.26: Offline AI Agents & Enterprise HandoverPraisonAI has released version 4.7.26 of its open-source Python agent framework, enabling fully offline AI agent execution. The update also introduces an early access feature for highly regulated teams, allowing AI tasks to be handed over to self-hosted models with a verifiable audit trail.