{ "specVersion": "1.1.0-draft", "note": "Qualitative scorecard. Grades mix core and optional capabilities and are not conformance certification.", "criteria": [ { "key": "zeroInstallUI", "label": "No terminal", "description": "Non-technical users can run skills and edit notes from a graphical app." }, { "key": "oneClickSOP", "label": "Portable skills", "description": "Markdown procedures with lifecycle and approval rules (Base); a launch dashboard and run history are optional Cockpit features." }, { "key": "byosAuth", "label": "Own subscription", "description": "Each user runs agents on their own AI plan through the official client, or their own API key." }, { "key": "teamSecondBrain", "label": "Second brain", "description": "Plain Markdown the owner controls, agent entry file, and tools to keep knowledge fresh." }, { "key": "teamRBAC", "label": "Permissions & safety", "description": "Authorization and untrusted-input rules (Base); enforced permissions per role belong to Team." }, { "key": "modularPlugins", "label": "Portability", "description": "Second brain as the single source of truth, harness and model switching, and a tools registry." } ], "implementations": [ { "name": "Kairos (formerly JARVIS, david-darr/kairos)", "category": "Sovereign Desktop AI Workspace & Vault UI", "harnessStatus": "Custom Bespoke Engine (No Open Harness)", "grade": "B+", "summary": "Self-hosted, open-source AI workspace with multi-model chat, automations, and a Markdown second-brain viewer. Rebranded from JARVIS to Kairos with the v2.0 release, featuring an editorial light UI, Forge coding environment, and native desktop installers.", "verdict": "Fits YAIBOS requirements for local file ownership and desktop delivery on Windows PC and Mac. It uses a custom runtime with its own maintenance needs.", "scores": { "zeroInstallUI": 5, "oneClickSOP": 3, "byosAuth": 4, "teamSecondBrain": 4, "teamRBAC": 1, "modularPlugins": 4 }, "strengths": [ "Packaged desktop releases for Windows PC, macOS, and Linux that do not require a terminal.", "Embedded Markdown vault graph visualization in Library directly inside the workspace.", "Chats, memory, notes, and automations live in local files you own." ], "gaps": [ "Not based on an open-source harness (such as DSH / Cordis): it uses a custom Python and Electron runtime, so it does not benefit from community harness plugins, fixes, or shared ecosystem improvements.", "Single-user desktop focus lacking multi-user team RBAC and folder-scoped permissions.", "Automations are not yet standardized into business skill cards with duration and ROI telemetry.", "Evolving plugin ecosystem under the new v2.0 architecture." ], "repoUrl": "https://github.com/david-darr/kairos", "screenshots": [ { "src": "https://yaibos.com/images/implementations/kairos-home.webp", "alt": "Kairos 2.0 Home: workspace dashboard with vault access, daily briefing, agents, and model status" }, { "src": "https://yaibos.com/images/implementations/kairos-tasks.webp", "alt": "Kairos 2.0 Tasks: automation routines, event triggers (webhooks/GitHub pushes), and execution history" }, { "src": "https://yaibos.com/images/implementations/kairos-vault.webp", "alt": "Kairos 2.0 Library Vault: interactive graph view of Markdown notes grouped by Projects, Resources, Personal, and Learning" } ], "securityAnalysisUrl": "https://yaibos.com/implementations/jarvis-security" }, { "name": "OBSIA (kevines-ods/OBSIA)", "category": "Obsidian Multi-Agent Harness & Skill Standard", "harnessStatus": "Multi-Harness Layer (Claude Code native / OpenCode / Aider)", "grade": "B+", "summary": "An open file-based specification for multiple agents. Agents, skills, scheduled tasks, and memory live in plain Markdown. It includes a modular installer that uses host context and a security contract for each agent.", "verdict": "The primary reference for YAIBOS file-based architecture. It runs skills, agent permissions, and memory on existing open CLI harnesses, keeping the files independent of a database.", "scores": { "zeroInstallUI": 2, "oneClickSOP": 3, "byosAuth": 5, "teamSecondBrain": 4, "teamRBAC": 2, "modularPlugins": 5 }, "strengths": [ "Agents, skills, and memory live entirely in an open Markdown second brain, viewed and edited with Obsidian.", "Per-agent RBAC security contract: declared read-only agents and bounded write zones, with unlisted mutations requiring human-reviewed Git patches.", "Context-aware modular installation: detects host tools (Docker, Proxmox, browser) to provision only relevant skills and agents, avoiding context window bloat.", "Native Claude Code integration, with portable prompt-generation scripts for other CLI harnesses (OpenCode, Aider, Goose)." ], "gaps": [ "Lacks a zero-install standalone web dashboard for non-technical team members (requires terminal or Obsidian setup).", "Permissions model is agent-level rather than multi-user team RBAC: currently designed for a single user/workstation.", "Scheduled tasks use versioned registries and timers, but lack 1-click UI execution buttons and duration telemetry.", "Early alpha release with documentation currently in French." ], "repoUrl": "https://github.com/kevines-ods/OBSIA" }, { "name": "DeepSeek Harness (dsh + Cordis)", "category": "Modular Agent Engine & Runtime", "harnessStatus": "Native Open Harness (Reference Runtime)", "grade": "B+", "summary": "An open-source agent harness built on the Cordis plugin engine, with swappable memory, tool, and execution plugins.", "verdict": "An architectural reference for a modular open-source agent harness, with community plugins. It currently lacks an interface for non-technical users and multi-user RBAC.", "scores": { "zeroInstallUI": 2, "oneClickSOP": 4, "byosAuth": 3, "teamSecondBrain": 3, "teamRBAC": 1, "modularPlugins": 5 }, "strengths": [ "Modular plugin architecture based on the Cordis meta-framework.", "Swappable open-source execution, memory (dsh-mnemon), and tool plugins.", "Execution telemetry, duration tracking, and audit logs." ], "gaps": [ "Strictly developer-oriented: requires terminal knowledge and Node.js to deploy.", "No native multi-user RBAC or folder permissions.", "Requires bridging for consumer subscription auth (ChatGPT/Claude web auth)." ], "repoUrl": "https://github.com/deepseek-ai" }, { "name": "Obsidian AI (sup3rus3r/obsidian-ai)", "category": "Visual Agent Canvas & Self-Hosted Platform", "harnessStatus": "Custom Visual Engine (No Open Harness)", "grade": "B+", "summary": "Self-hosted, open-source platform for orchestrating AI agents through a visual drag-and-drop canvas with native MCP support, prompt vault, and team RBAC.", "verdict": "Self-hosted visual orchestration with secrets management and support for several LLM providers. Its node canvas is more complex than one-click business skill buttons, and it lacks a native Markdown editor that reads and writes notes.", "scores": { "zeroInstallUI": 4, "oneClickSOP": 4, "byosAuth": 3, "teamSecondBrain": 3, "teamRBAC": 4, "modularPlugins": 4 }, "strengths": [ "Drag-and-drop visual canvas for agent teams and DAG execution workflows.", "Production security: Fernet secrets vault, AES encryption, JWT auth, and role-based access control.", "Native Model Context Protocol (MCP) support and multi-provider model switching without config changes." ], "gaps": [ "Not built on a standard open harness like DSH or OpenCode: uses a bespoke node-graph runtime, meaning it cannot natively run community harness plugins or shared skill workflows.", "Node-and-wire visual canvas still demands workflow logic design rather than simple 1-click skill cards for business staff.", "Does not provide an integrated, human-editable Markdown second-brain note vault.", "Self-hosting requires Docker or server deployment before non-technical staff can use it." ], "repoUrl": "https://github.com/sup3rus3r/obsidian-ai" }, { "name": "LoopSuit AI", "category": "Custom AI Systems & Automation Studio", "harnessStatus": "Bespoke Agency Service (No Open Harness)", "grade": "B", "summary": "AI-native product studio and automation agency deploying custom autonomous agents, onboarding pipelines, and bespoke operational workflows for business operators.", "verdict": "Custom agency deployments with a focus on operational design and rapid delivery of production-ready MVPs. It does not provide a self-hosted, open-source second-brain operating system.", "scores": { "zeroInstallUI": 4, "oneClickSOP": 4, "byosAuth": 2, "teamSecondBrain": 2, "teamRBAC": 2, "modularPlugins": 3 }, "strengths": [ "Custom autonomous agent workflows for lead intake, client onboarding, and operational procedures.", "Product design and two-week delivery for operators.", "Clients own the deployed automation system." ], "gaps": [ "Not an open harness and not built on open runtimes like DSH or OpenCode: custom boutique agency automation built on private client infrastructure.", "Bespoke agency delivery rather than a standardized, open-source sovereign software harness.", "No integrated, bi-directional Markdown second-brain knowledge layer.", "Auth and pricing models vary per client setup rather than standardized BYOS session auth." ], "repoUrl": "https://loopsuitai.com" }, { "name": "OpenDesign / DSHKer", "category": "Visual Harness UI", "harnessStatus": "Built on Open Harness (DeepSeek Harness)", "grade": "B", "summary": "A local-first visual desktop interface built on DeepSeek Harness (DSH), designed to make the harness accessible to non-technical users.", "verdict": "Uses DSH's modular plugins, execution telemetry, and community updates through a visual desktop layer that does not require terminal setup.", "scores": { "zeroInstallUI": 4, "oneClickSOP": 4, "byosAuth": 2, "teamSecondBrain": 2, "teamRBAC": 1, "modularPlugins": 4 }, "strengths": [ "Directly powered by DeepSeek Harness (DSH), inheriting upstream community plugins and runtime improvements.", "Zero-install desktop app for non-technical team members.", "Visual preview of agent executions and artifacts." ], "gaps": [ "No integrated team second brain or Markdown editor.", "Single-user desktop focus with zero role-based permissions.", "Early stage with evolving API stability." ], "repoUrl": "https://github.com/nexu-io/open-design", "screenshots": [ { "src": "https://yaibos.com/images/implementations/open-design.webp", "alt": "Open Design home screen: prompt box with output types (slide deck, UI mockup, document) and a model picker" } ] }, { "name": "Khoj", "category": "AI Second Brain & Assistant", "harnessStatus": "Not an Agent Harness (Conversational Search)", "grade": "B-", "summary": "Self-hosted AI search and personal assistant that indexes local Markdown notes and documents.", "verdict": "Strong on local Markdown second-brain capabilities, but lacks an operational 1-click business skill runner.", "scores": { "zeroInstallUI": 4, "oneClickSOP": 2, "byosAuth": 2, "teamSecondBrain": 4, "teamRBAC": 2, "modularPlugins": 3 }, "strengths": [ "Local Markdown indexing and semantic retrieval.", "Clean web and desktop interface.", "Self-hosted privacy guarantees." ], "gaps": [ "Not an autonomous agent harness and not built on open harnesses like DSH or OpenCode: custom Python engine focused on personal document retrieval and conversational search.", "Oriented toward conversational Q&A rather than running business skills.", "No 1-click action cards with telemetry and duration tracking.", "Rudimentary team sharing and permission controls." ], "repoUrl": "https://github.com/khoj-ai/khoj" }, { "name": "Obsidian (with Agent & Smart Plugins)", "category": "Knowledge Base + Plugin Ecosystem", "harnessStatus": "Fragmented Plugins (No Unified Harness)", "grade": "B-", "summary": "A local Markdown editor with community agent plugins (Smart Connections, Banners, Local LLMs).", "verdict": "Suited to editing a personal second brain. Setup is complex for non-technical staff, and it lacks multi-user RBAC.", "scores": { "zeroInstallUI": 2, "oneClickSOP": 2, "byosAuth": 2, "teamSecondBrain": 5, "teamRBAC": 1, "modularPlugins": 4 }, "strengths": [ "Plain Markdown files remain under your control.", "A large plugin ecosystem and local graph visualization.", "Zero proprietary data format lock-in." ], "gaps": [ "Lacks a cohesive open harness foundation like DSH or OpenCode: community plugins operate as isolated silos with inconsistent tool permissions and fragmented context.", "High onboarding friction for non-technical team members.", "Obsidian Sync has zero folder-level RBAC for team permissions.", "Agent execution relies on fragmented community plugins rather than a cohesive runtime." ], "repoUrl": "https://obsidian.md" }, { "name": "AnythingLLM", "category": "Desktop AI Workspace", "harnessStatus": "Not an Agent Harness (Document RAG)", "grade": "C+", "summary": "All-in-one desktop and enterprise app for chatting with documents and local models.", "verdict": "Polished document workspace, but strictly conversational with no concept of business skill execution.", "scores": { "zeroInstallUI": 5, "oneClickSOP": 1, "byosAuth": 3, "teamSecondBrain": 3, "teamRBAC": 3, "modularPlugins": 2 }, "strengths": [ "One-click desktop installer with zero technical setup required.", "Supports multi-user workspaces and basic permission levels.", "Flexible model endpoint configuration." ], "gaps": [ "Not an agent harness and not built on open runtimes like DSH or OpenCode: custom monolithic Node.js backend designed for document Q&A, lacking autonomous business skill execution.", "No 1-click business skill procedures or automated business agent loops.", "Data locked in internal vector databases instead of plain Markdown.", "Rigid plugin architecture with limited extensibility." ], "repoUrl": "https://github.com/Mintplex-Labs/anything-llm" }, { "name": "Dify / Flowise", "category": "Visual Workflow & LLMOps", "harnessStatus": "Custom Visual DAG Engine (No Open Harness)", "grade": "C", "summary": "Visual canvas builders for chaining LLMs, tools, and multi-agent workflows into APIs.", "verdict": "Visual node workflows suit developers but are complex for non-technical teams. They lack owner-controlled Markdown notes.", "scores": { "zeroInstallUI": 3, "oneClickSOP": 3, "byosAuth": 2, "teamSecondBrain": 1, "teamRBAC": 3, "modularPlugins": 3 }, "strengths": [ "Powerful visual logic builder for custom multi-agent chains.", "Good team access management in enterprise self-hosted tiers.", "Wide support for external API connectors." ], "gaps": [ "Uses proprietary visual node execution graphs rather than an open harness runtime like DSH or OpenCode, creating workflow lock-in within their specific visual schemas.", "Visual node graphs are intimidating for non-technical operators.", "Relies on PostgreSQL and cloud vector stores, completely missing sovereign Markdown files.", "Heavy infrastructure footprint to self-host." ], "repoUrl": "https://github.com/langgenius/dify" }, { "name": "Open WebUI", "category": "Self-Hosted Model Interface", "harnessStatus": "Not an Agent Harness (Model Chat UI)", "grade": "C+", "summary": "A self-hosted chat interface for Ollama, OpenAI, and compatible endpoints.", "verdict": "A self-hosted alternative to ChatGPT, without an operational skill catalog or agent action loops.", "scores": { "zeroInstallUI": 4, "oneClickSOP": 1, "byosAuth": 3, "teamSecondBrain": 2, "teamRBAC": 3, "modularPlugins": 3 }, "strengths": [ "Familiar and clean ChatGPT-style user interface.", "Good user management and multi-role RBAC in server mode.", "Broad model support including local Ollama runners." ], "gaps": [ "A web chat interface for LLMs rather than an agent harness: not built on DSH or OpenCode, and lacks an autonomous task loop or business skill runner.", "Purely conversational: no 1-click execution cards or task duration telemetry.", "Lacks a bi-directional Markdown second brain.", "Not designed to automate repetitive business processes." ], "repoUrl": "https://github.com/open-webui/open-webui" }, { "name": "CrewAI / AutoGPT / OpenManus", "category": "Developer Agent Frameworks", "harnessStatus": "Code Frameworks (Custom Runtimes)", "grade": "D+", "summary": "Code-first multi-agent orchestration libraries (CrewAI, AutoGPT, OpenManus) providing programmatic agent harnesses for developers.", "verdict": "Support algorithmic experiments through their own custom runtimes. They do not build on modular standards such as DSH or OpenCode, and lack an interface for non-technical users.", "scores": { "zeroInstallUI": 1, "oneClickSOP": 2, "byosAuth": 1, "teamSecondBrain": 1, "teamRBAC": 1, "modularPlugins": 4 }, "strengths": [ "Flexible multi-agent role delegation in Python or TypeScript.", "Rapidly adopting open standards like Model Context Protocol (MCP)." ], "gaps": [ "Each library invents its own custom programmatic harness rather than converging on modular standards like DSH (Cordis) or terminal standards like OpenCode, resulting in framework churn and developer lock-in.", "Zero graphical UI: requires terminal execution and virtual environments.", "Metered API use can produce unexpectedly large bills.", "No persistent human-readable knowledge layer or team permissions." ] } ] }