Your team's AI.
One governed source of truth.
One governed vault your whole team's AI answers from — approved, versioned, self-hosted. Deploy free in one command.
The problem: Context fragmentation.
Right now, every person on your team is feeding their AI something different. Your engineers have their own context files. Your ops manager uploads PDFs directly into Claude. Your newest hire is going off the wiki, which nobody's touched since Q3.
The result: the same question gets five different answers. Your AI cites the pricing sheet you deprecated six months ago. A decision got documented and nobody's model knows about it.
The most valuable context of all lives only in people's heads — and walks out the door when they do.
ContextNest Community Edition puts it somewhere governed. Your team's living knowledge becomes a self-hosted, versioned vault with a cryptographic hash chain behind every document. Every query is traceable. Before any document reaches your AI, it passes through steward approval — so only current, approved content feeds your model.
The result: everyone's AI draws from the same approved, versioned knowledge — not from whatever ended up in their last chat window.
Versioned
Every document change is hash-chained and tamper-evident. The complete history is always reconstructible.
Governed
Documents require steward approval before reaching your AI. Draft content stays in draft — it never feeds a model.
Auditable
Every query produces a complete trace: what your AI consumed, from which version, at what time.
How ContextNest works.
Three steps from raw knowledge to governed AI context.
Structure your knowledge
Write in Markdown. Use [[wiki links]] to connect related concepts. Tag documents with #topics. Add YAML frontmatter to define type, author, status, and relationships. ContextNest turns your documents into a navigable knowledge graph — not a flat file dump.
--- title: Q3 Pricing Strategy type: document tags: [#pricing, #strategy] status: draft --- We are shifting focus toward the [[Enterprise-Tier-Pricing]] model proposed by @Sarah.
Govern what reaches your AI
Documents don't reach your AI automatically. In governed mode, every document passes through steward approval before it's compiled into the active context. Draft content stays in draft. Outdated content gets versioned, not deleted. Your AI only works from what's been explicitly approved.
Connect your AI
Once approved, your context is queryable via the Model Context Protocol (MCP). Any MCP-compatible AI client — Claude, Cursor, Claude Code — connects to your vault directly. No custom integrations. No re-uploading. No context window paste-and-pray.
{
"mcpServers": {
"contextnest": {
"command": "node",
"args": ["node_modules/@promptowl/contextnest-mcp-server/dist/index.js"],
"env": {
"CONTEXTNEST_VAULT_PATH": "/path/to/your/vault"
}
}
}
}Built for the engineer. Governed for the org.
Whether you're the engineer deploying it or the organization running it, ContextNest balances developer sovereignty with strict content governance.
| Feature | Ben (Engineer / Self-Hosted) | Team / Small Org |
|---|---|---|
| Self-Hosted | His data stays on his infra. No vendor dependency. | Company IP never goes to a third-party SaaS. |
| SHA-256 Hash Chains | Tamper-evident, reconstructible version history. | Every change is logged. Nothing is silently overwritten. |
| Steward Approval | Proof that the governance claim is cryptographically real. | AI only answers from what's been explicitly approved. |
| MCP-Native | Connects to Claude, Cursor, and any compliant client. No custom integrations. | Every teammate connects their favorite editor without IT friction. |
| Collaborator Access | Multi-user structure meets engineering compliance specs. | Add teammates with read/write access per person or per tag. |
| External-Edit Detection | Catches raw git or filesystem edits, showing instant diffs. | If someone bypasses the UI, the changes surface instantly for review. |
| Inline Version Diff | Full audit trail. Reconstructible state at any checkout. | Visually inspect exactly what changed between approved versions. |
The governance model, in depth.
Documents require approval before they reach your AI. Draft content stays in draft — permanently — until a steward approves it. Stewardship, scope resolution, review queues, and audit log streaming are built into Community and Enterprise editions.
Deploy the Governed Community Server in three steps.
Start the self-hosted community server for team collaboration and stewardship. To run ContextNest as an independent, single-user developer tool instead, see the local CLI and Desktop options below.
Deploy Server
Run the community server on your machine or VPS using npx.
Activate License
Open http://localhost:3838 in your browser. Paste your free Community License key from your PromptOwl account to activate.
Connect Your Team's AI
Add this MCP config to Claude Desktop or Cursor settings to query approved knowledge.
{
"mcpServers": {
"contextnest": {
"command": "node",
"args": [".../contextnest-community/dist/mcp.js"],
"env": { "CONTEXTNEST_API_KEY": "cnst_..." }
}
}
}Run ContextNest natively from the CLI.
One npm install gives you the ctxcommand anywhere — a local, version-controlled vault every AI assistant on your machine can read. No server, no license key.
Install the CLI, create a vault, connect your agents.
The Claude Code plugin.New
Working in Claude Code, the plugin handles the vault for you — relevant material is pulled into each prompt automatically, and new facts, decisions, and gotchas are captured back at the end of every turn. It runs through the ctxCLI, so there's no MCP wiring to set up.
On enable, four optional settings: retrieval effort (off · search · query · agent), auto-capture (on by default), a pinned vault alias, and the ctx binary override. Leave the vault empty and the agents pick the relevant vault automatically.
Live query tools via MCP.
For runtime access instead of config files, add the ContextNest MCP server so agents can resolve, search, and read your vault live. Point it at your vault directory:
{
"mcpServers": {
"contextnest": {
"command": "npx",
"args": ["-y", "@promptowl/contextnest-mcp-server", "/absolute/path/to/your/vault"]
}
}
}No terminal required.
ContextNest Desktop gives you the same local governed vault without CLI commands or configuration. Download for macOS, Windows, or Linux.
macOS
Apple Silicon (M1–M4) by default — Intel build available too
- macOS 13 Ventura or later
- Apple Silicon (M1/M2/M3/M4) or Intel
- Minimum 8GB RAM
Windows (64-bit)
Standard installer (x64) — portable build available too
- Windows 10 or 11 (64-bit)
- Intel Core i3 or AMD equivalent
- Minimum 8GB RAM
Linux
Universal AppImage build — portable archive available too
- Ubuntu 22.04+, Fedora 37+ or equivalent
- 64-bit (x86_64), glibc 2.31 or newer
- Minimum 8GB RAM
Ready-to-use prompts.
Copy these directly into Claude, Cursor, or any AI. They work better once ContextNest is installed and your vault is connected.
Set up a new project context
Query your vault
Capture a decision
Onboard to a project
Build a starter vault
Don't start from scratch.
Pick a vault template and give your AI instant, structured context.
Executive / Leadership
Strategy, operations, and leadership playbooks for senior leaders.
Developer / Engineering
Architecture decisions, coding standards, and fast onboarding for engineering teams.
Sales / Revenue
Objection handling, competitive intel, and enablement playbooks for sales teams.
Questions, answered.
What is ContextNest?
ContextNest is wiki-style persistent memory for AI models — patent pending. It structures your knowledge into versioned, graph-linked, cryptographically verified Markdown documents that AI agents read directly via the Model Context Protocol (MCP). Your AI gets permanent, governed context — it works from what you actually know, not from similarity guesses or expired documents.
How is this different from RAG or a vector database?
ContextNest handles organic, living knowledge — the policies, processes, and decisions that evolve. Documents are structured, versioned, and human-approved, and your AI reads them directly — no embeddings, no similarity search — so you get the same answer every time. RAG and vector databases handle static knowledge: semantic search over large corpora that rarely change. Different jobs, and PromptOwl governs both.
How do we control what our AI answers from?
Steward approval. Documents don't reach your team's AI until a steward approves them — draft content stays in draft, outdated content gets versioned rather than deleted, and every change lands in a cryptographic hash chain. Every query produces a complete trace: what the AI consumed, from which version, at what time.
Where does our data live?
On your infrastructure. Your vault is a directory of Markdown files on your own machine or VPS — no data leaves your environment, no external API calls from the governance layer, no vendor dependency. Move it, back it up, take it with you.
Does ContextNest fix AI hallucinations?
Partially. ContextNest fixes the governance failure mode — AI citing outdated, unapproved, or incorrect content because the context layer had no governance. It does not fix model-level hallucinations where the model invents facts not present in any document. What you get: an AI that works only from approved, current, auditable knowledge.
What does it cost?
Community Edition is free for self-hosting — a free PromptOwl account issues your license key, and you govern your team's context without licensing fees. ContextNest Desktop is free for individual use. PromptOwl Platform and Enterprise tiers add hosted options, agents, white-labeling, SSO, and full compliance tools.
What AI tools does ContextNest work with?
Any MCP-compatible AI client — Claude Desktop, Claude Code, Cursor, Antigravity. Claude Code users can also install the ContextNest plugin (/plugin install contextnest), which adds automatic retrieval from your vault and automatic capture of new knowledge at the end of each turn — no MCP wiring required.
Your team's AI deserves a
governed source of truth.
Free. Self-hosted. One PromptOwl account to activate. Deploy in minutes, connect your team, and stop getting five different answers to the same question.