# Kw · Knowledge Wiki — element 15 of 58

> The company's shared brain. Turns scattered context into a substrate agents can read.

- **Group:** 4 · Knowledge & Memory
- **Necessity:** Essential
- **Price band:** $ · under $30/mo
- **Maturity:** Stable
- **Edition:** v2026.Q3 · verified 2026-09-13

## Leading tools (v2026.Q3)

- **Notion** — the default company brain
- **Obsidian** — plain files, agent-native
- **Slite** — verified answers, deepest mcp
- **Outline** — self-hostable, mcp built-in
- **Superhuman Docs (ex-Coda)** — docs that compute

## Our take

Your wiki is no longer for humans only. Write it so an agent can act on it — that's the 2026 shift.

## Combines with

Mc, Vd, Mt
- Guide: [Best knowledge base for AI agents (2026)](https://elems.ai/best/best-knowledge-base-for-ai-agents.html)

## The top 5 — deep dossier (verified 2026-09-13)

Notion, for most teams — 100M users, $500M ARR, an $11B valuation, and since 3.0 (Sep 2025) the wiki, the agents that act on it, and a hosted MCP server live in one product at $20/seat. Obsidian is the pick when you want the substrate itself to be agent-native — plain Markdown files any coding agent reads without an API, now with official agent skills from the CEO himself. Slite wins when trustworthy answers matter more than breadth: doc verification plus the deepest purpose-built MCP (40+ tools). Outline when you need self-hosting or flat pricing; Superhuman Docs (ex-Coda) when your wiki needs to compute — tables, automations, 800+ integrations — and you can stomach the rebrand turbulence.

1. **Notion** (Notion Labs) — Free · Plus $10/seat/mo · Business $20 (full AI + Agent) · Enterprise custom · Custom Agents $10 per 1,000 credits. Best for: Teams that want the company brain, the AI that answers from it, and the agents that maintain it in one workspace. Why: The scale leader — 100M users (2024), $500M ARR (Sep 2025), $11B valuation (Jan 2026 tender) — and the furthest along on agent-readability: Notion 3.0 (Sep 18, 2025) made agents first-class, Custom Agents run on schedules/triggers since Feb 2026, and the hosted MCP server gives any external agent OAuth read/write in one click. The wiki that agents both consume and garden. Watch: Full AI is gated to the $20 Business plan, and Custom Agents added usage-based credit billing on May 4, 2026 — costs are unpredictable at agent scale. Your knowledge lives in their cloud; Markdown export is lossy for databases. Agent sessions cap at ~20 minutes. [https://www.notion.com](https://www.notion.com)
2. **Obsidian** (Dynalist Inc. (Obsidian)) — Free (personal, no limits) · Commercial $50/user/yr · Sync $4/mo · Publish $8/mo. Best for: Founders and engineering-heavy teams who want knowledge as plain local files that any coding agent can read, grep, and edit — no API, no vendor. Why: The most agent-native substrate in the category: a vault is just Markdown on disk, so Claude Code, Codex, or any file-capable agent operates on it directly — put it in git and you get versioned, reviewable knowledge. CEO Steph Ango shipped official open-source agent skills (Obsidian Markdown, Bases, JSON Canvas, CLI) in June 2026 — 37k GitHub stars within weeks — codifying the vault conventions agents should follow. Watch: Not a team wiki out of the box: no browser editing, no granular permissions, collaboration means shared Sync vaults. No hosted product, no admin controls — governance of the company brain is entirely DIY. The app itself is not open source. [https://obsidian.md](https://obsidian.md)
3. **Slite** (Slite) — Basic $10/user/mo · Pro $20 (agent + 12+ connected sources) · Enterprise custom · MCP + API on all paid tiers. Best for: Teams who want a dedicated knowledge base whose answers agents can actually trust — verification workflows plus the deepest wiki MCP. Why: The most complete agent-readability story of any pure knowledge base: a remote MCP server (api.slite.com/mcp, OAuth, no config) exposing 40+ tools for search, doc read/write, and structured tables, with permission inheritance — works with Claude, ChatGPT, Cursor, Gemini, and custom frameworks. Doc verification and AI fact-checking with suggested fixes attack the real failure mode of company wikis: stale pages that agents confidently repeat. Watch: A small independent vendor next to Notion and Atlassian — platform-longevity risk is real. Basic caps AI at 30 questions/seat/mo; the agent and cross-tool search need the $20 Pro tier, at which point it prices like Notion Business with a fraction of the ecosystem. [https://slite.com](https://slite.com)
4. **Outline** (General Outline (BSL 1.1)) — Cloud flat-rate: $10/mo (≤10 users) · $79/mo (≤100) · $249/mo (≤200) · self-hosted free. Best for: Teams that want a fast Markdown wiki they can self-host — or cloud pricing that stops scaling per seat. Why: The self-hostable choice with real momentum: 39.7k GitHub stars, source-available (BSL 1.1), and MCP built into every workspace since Feb 18, 2026 — agents search, read, create, and edit docs natively, with an expanded toolset (patching, moves, attachments, comments) shipped Apr 26, 2026. Flat-rate cloud pricing ($79/mo for up to 100 people) undercuts every per-seat rival at team scale. Watch: BSL is not OSI-open-source — check the license terms if that matters. In-product AI is thin (Q&A only) — it's the substrate for external agents, not an agent platform. Small commercial team behind the cloud offering. [https://www.getoutline.com](https://www.getoutline.com)
5. **Superhuman Docs (ex-Coda)** (Superhuman (ex-Grammarly; acquired Coda Dec 2024)) — Free · Pro $10/Doc Maker/mo · Team $30/Doc Maker · Enterprise custom — editors and viewers free. Best for: Teams whose wiki needs to compute — living docs with tables, buttons, automations, and two-way syncs, not just prose. Why: Coda's doc-as-app engine inside the Superhuman suite: knowledge hubs with structured tables, workflow automation, 800+ integrations, and MCP support that lets ChatGPT, Claude, and Cursor read and act on docs. The Doc Maker pricing model (only builders pay; editors and viewers free) stays the cheapest way to give a whole company access to structured knowledge. Watch: Brand and roadmap turbulence: coda.io now redirects to superhuman.com, and the product competes for attention inside a suite (Mail, Grammarly, Go) whose center of gravity is not wikis. Doc-Maker promotions quietly multiply bills; Packs can carry separate fees. [https://superhuman.com/docs](https://superhuman.com/docs)

### How to choose
- If You want one product where the wiki, the search, and the agents that maintain it all live together → Notion Business at $20/seat — accept the credit-billing line item for Custom Agents and watch it monthly; usage pricing arrived May 4, 2026.
- If Your team already lives in coding agents (Claude Code, Codex) and treats knowledge as code → An Obsidian vault in a git repo, with the official obsidian-skills conventions — agents read and write it like any other files, with PR review as your governance layer.
- If Agents answering from stale docs scares you more than missing features → Slite — verification workflows plus fact-checking keep the substrate trustworthy, and its 40+-tool MCP is the deepest of any dedicated knowledge base.
- If Compliance requires self-hosting, or per-seat pricing offends you at 50+ people → Outline — free on your own infra, or $79/mo flat for up to 100 users in cloud, with MCP built into every workspace since Feb 2026.
- If Procurement is Atlassian-shaped or the org is 1,000+ people → Confluence with Rovo — the incumbent nobody gets fired for, with an MCP server included on every cloud plan (rate-limited to 500–10,000 calls/hr by tier).

### The field (22 more)

Confluence + Rovo, Slab, Anytype, Capacities, Guru, Superhuman Go / suite, Tana, Logseq (fading), Roam Research (fading), Mem, Craft, Heptabase, Reflect, AFFiNE, AppFlowy, Nuclino, Tettra, BookStack, Wiki.js (fading), Almanac (dead), Dropbox Paper (fading), Coda (renamed)

Full dossier data: https://elems.ai/e/kw.json

---
Source: [elems.ai](https://elems.ai/e/kw.html) — the periodic table of the AI-led startup. Data: https://elems.ai/elements.json (CC BY 4.0, cite elems.ai).
