Compare
Where KAOS fits.
KAOS gets compared against three different groups, and the comparison is different in each one. The proprietary legal-AI platforms (Harvey, CoCounsel, Lexis+ Protégé, Westlaw Advantage, Everlaw, Luminance, Ironclad Jurist) sell finished workflows hosted in their cloud. The open-source agentic frameworks (DSPy, LangChain, LangGraph, Pydantic-AI, Mirascope, Instructor) sell developer toolkits with no legal grounding. And a new wave of open-source legal-AI apps (Mike OSS, OpenContracts, LQ.AI, doc.haus) ships finished, self-hostable tools — KAOS is the composable layer underneath them.
Three axes, three groups
Where each group lands.
The detailed tables below score each named vendor on a long list of capabilities. Stepping back: only one of the three groups scores on all three of open source, MCP-native by construction, and legal-domain grounding at once.
Reading row by row: KAOS scores on all three axes by construction. The proprietary platforms ship deep legal workflows but stay closed and have no native MCP surface. The OSS frameworks are open and Pythonic but were not built around legal documents, citations, or recipes — and they sit on the consumer side of MCP, not the server side. A fourth group, the open-source legal-AI apps, clears all three axes too, so the real question there is different: adopt a finished app, or build on the substrate beneath it. That table is last.
Against proprietary platforms
Open source where they are closed.
The incumbents sell finished workflows hosted on their cloud, with the data inside their walls. KAOS ships the building blocks (open-source, self-hostable, served over the Model Context Protocol) that a firm or a vendor can assemble into the same workflows on its own terms.
| Capability | KAOS | Harvey | CoCounsel | Lexis+ | Westlaw | Everlaw | Luminance | Ironclad |
|---|---|---|---|---|---|---|---|---|
| Open source | Apache-2.0 | No | No | No | No | No | No | No |
| MCP-native | Yes — to_mcp_dict() built in | No | No | No | No | No | No | No |
| Provenance to source span | Yes — page + bbox + char_span | Yes | Yes | Yes | Yes | Yes | Yes | Yes |
| Refusal when uncertain | GroundedAnswer[T] schema | Confidence-backed | Confidence-backed | Confidence-backed | Document Analyzer flags | Yes — gold standard | Yes | Yes |
| Citation verification VLAIR Capability 3 | kaos-citations · 8 kinds, 8 resolvers, NLI | Citation checker | Yes | Shepard's Citation Agent | Litigation Document Analyzer | Yes | Yes | Yes |
| EDGAR research workflows VLAIR #7 | kaos-source — direct API | Deep Research | Yes (federated) | Workflow builder | SEC integration | Partial | Compliance scans | No |
| Bring your own corpus | Yes — user owns data | Vault | Federated | Workflows | Limited | eDiscovery only | Yes | CLM only |
| Self-host anywhere laptop to cloud | Yes — pip install | No | No | No | No | No | No | No |
| Customer can extend with new tools | Yes — write a KaosTool | Workflow Builder (no code) | Custom workflows | Custom workflow builder | Limited | No | No | No |
Against open-source frameworks
Legal grounding where they are general.
The OSS frameworks Python developers reach for first. KAOS is closer in spirit to DSPy than to any other; the head-to-head is on the LLM surface page.
| Capability | KAOS | DSPy | LangChain | LangGraph | Pydantic-AI | Mirascope | Instructor |
|---|---|---|---|---|---|---|---|
| Typed Signatures / Programs | Yes — DSPy-descendant | Yes — original | Partial | Partial | Yes | Yes | Output-only |
| Optimizer surface | 10 (Bootstrap, MIPROv2, …) | 14+ (deepest) | No | No | No | No | No |
| MCP server out of the box | Yes — kaos-mcp + 14 servers | Tool consumer only | Tool consumer only | Tool consumer only | No | No | No |
| AST-grounded outputs | Cited[T] · Answer[T] · GroundedAnswer[T] | No | Document loaders | Document loaders | No | No | No |
| Crash-safe batch with workspace | Yes — Program v3 envelope + SQLite | Evaluate (eval-shaped) | Manual | Manual | Manual | Manual | Manual |
| Recipes for legal work | Yes — 11 named legal recipes | No | Generic templates | Generic templates | No | No | No |
| Citation extraction + verification | kaos-citations · 8 kinds + NLI | No | No | No | No | No | No |
| Document AST with provenance | kaos-content · 34 node types | No | String chunks | String chunks | No | No | No |
| Rust+PyO3 BM25 / tokenizer / NLP routines | kaos-nlp-core · <600 µs / 2-term | No | Python BM25 | Python BM25 | No | No | No |
Against open-source legal AI
Infrastructure where they are applications.
A new wave of open-source legal AI ships finished, self-hostable chat-and-review apps — Mike OSS, OpenContracts, LQ.AI, and doc.haus are the most active. Unlike the proprietary platforms, most are permissively licensed and clear the same open-source and legal-grounding bars KAOS does. The difference is what you get: KAOS is the typed documents, provenance, and MCP tools an app like these is built from, not a finished app itself.
@273v/kaos-ui-react (alpha) gives you drop-in React chat, transcript, and citations-panel components on npm, and kaos-ui scaffolds a wired React, Streamlit, or Textual starter in one command. Pick KAOS when you are building your own legal-AI product or embedding into systems you already run, and want composable packages, a typed importable document AST, and MCP-native tools as the layer underneath. Two licensing notes: Mike OSS is AGPL-3.0, whose network-copyleft can extend to modified versions you distribute or offer as a service; OpenContracts, LQ.AI, and KAOS are permissive.
| Capability | KAOS | Mike OSS | OpenContracts | LQ.AI | doc.haus |
|---|---|---|---|---|---|
| License | Apache-2.0 — permissive | AGPL-3.0 — copyleft | MIT — permissive | Apache-2.0 — permissive | MIT — permissive |
| What it is | 18 composable Python packages | Finished web app | Platform + API | Self-hosted app | Finished agent app |
| Self-host laptop to cloud | Yes | Yes | Yes | Yes | Yes |
| Usable by a non-developer today | No — scaffold a starter, then build | Yes — finished UI | Yes — deploy & use | Yes — finished UI | Yes — finished UI |
| Finished review UI tabular / approve-reject | No — components, not a finished grid | Tabular review | Per-cell approve/reject grid | Playbooks + tabular review | Q&A + DOCX redlines |
| Ready-made UI you can import npm components + scaffolder | Yes — @273v/kaos-ui-react (chat, citations panel, doc explorer) + kaos-ui scaffolder | In-app only | In-app only | In-app only | In-app only |
| Extra privacy controls beyond self-hosting | Self-host — you set the boundary | Encrypted matter vault | Self-host | Anonymization layer for cloud inference | Text + embeddings never leave disk |
| Install one piece via pip | Yes — pip install kaos-pdf | No — deploy the app | No — deploy the platform | No — deploy the stack | No — clone & run |
| Public, typed AST you can import | kaos-content · 34 node types, page+bbox+char_span | No — internal to the app | No — annotations stay in-platform | No — internal to the app | No — edits files directly |
| Citation grounding / verification | kaos-citations · substring + NLI | CourtListener case-law | Span-anchored to source PDF | Four-stage, character-verifiable | Clause-cited answers |
| MCP | Native — 14 servers, 200+ tools | No | Built-in MCP server | MCP connector client | No |
| Use as a library / building block | Yes — it is the substrate | No — finished app | Via its API | Open, forkable skills (agentskills.io) | No — finished app |
Compiled from each project's public repository and site, June 2026: Mike OSS (AGPL-3.0), OpenContracts (MIT), LQ.AI (Apache-2.0), doc.haus (MIT). General-purpose open agentic apps with legal use-case blogs (Eigent, ibl.ai) are a different category. This field moves fast — spotted a stale cell? Open an issue.
The lesson the market just taught
A thin layer over a model API is not a durable business.
In late 2025, Robin AI, a well-funded contract-review startup, sold its managed services to Scissero; the engineering team was acquired by Microsoft in January 2026. The lesson the market drew: durable value lives in workflow depth, in answers traceable back to evidence, and in the work of integrating with real systems. Access to the model is not enough. Thin contract-review wrappers could not survive once the underlying model APIs caught up.
KAOS is the open-source answer. Every answer traces back to a page and a bounding box. Every dependency has been license-audited, with no AGPL exposure. The whole platform speaks the Model Context Protocol, so Claude Code, Codex CLI, Gemini CLI, VS Code, and Cursor can call it directly. The recipes are named for real legal work: merger agreements, leases, court opinions, EDGAR research. Not chatbot demos.
See /why-kaos for the architectural argument and /runtime, /extraction, /llm, /agentic, /legal-intelligence, and /search-and-retrieval for what each layer does.
On these comparisons. Competitor capabilities are compiled from public vendor documentation and product pages as of June 2026; vendors ship quickly, so treat each cell as a pointer to verify, not a fixed score. KAOS rows are verified against the published source at github.com/273v. Spotted something out of date? Open an issue and we'll correct it.