“Two of our agents told people different things.”
Why it happensEach agent has its own copy of the instructions and documents. The copies drift apart, and neither agent knows the other exists.
Usually starts at the second agent
Every AI your team uses should already know your business — and you should be able to prove what it knows.
Today your AI learns from a drive, a wiki or a pasted brief that no one approved. ContextNest gives every AI the same approved company knowledge and records every read, so you can rebuild what it knew on any date.
Example
One person, one agent, their own copy of the policy. It works, because it's theirs.
Works with the AI tools you already have.
Teams and partners using PromptOwl
Where it breaks
Your AI worked when it was yours. It breaks when it becomes everyone's.
It starts with one person getting AI to work for them. The trouble starts the day it has to be shared: the same knowledge, pulled in different directions.
The symptoms
All the symptoms, and the moments they start →Find the one your team has said out loud.
“It was brilliant on my laptop and dumb in production.”
Why it happensOn your laptop the agent had your logins, your files, your history and no time limit. In production it gets narrow credentials, a time limit and none of what it had learned.
Starts when it leaves one machine
“An agent read something it was told to stay out of.”
Why it happensWriting “don't touch client data” in a prompt is a request. If the agent can reach the folder, it will open it eventually and apologize afterward.
Critical once agents face clients
What you'll try
You've probably tried two of these already.
Each one is good at what it was built for. Each one strains at a different point once AI is shared across people, agents and models.
Point the agent at the files
The shared drive or SharePoint
Where it breaks: Search gets worse as files pile up, old versions answer questions, and an agent using your connector can open everything you can.
Full comparison →Keep it in the assistantClaude or ChatGPT Projects
Where it breaks: Great for one person. Tied to one vendor and one account, and it leaves when that person leaves.
Full comparison →Plug memory into the agentA memory tool
Where it breaks: Quick to add to one app. The agent writes its own memories, and nobody approves what it keeps.
Full comparison →Why teams pick us
Context and governance that just works.
When it breaks, it's rarely the model. It's what the model was given. ContextNest governs that while you work. You write, chat and run agents, and the approvals, versions and read records happen along the way.
- Runs anywhere. Same nest.
- Start with plain files on a laptop with the open-source CLI, self-host the server in Docker, or let us host it. Push the same nest from one to the next, with no migration.
- No infrastructure to stand up
- No graph database, no vector cluster, no data platform, no ontology pipeline. It starts as plain files, and hosted is free to start.
- A product people open every day
- A Notion-style wiki editor, approvals, version history and compare, a read log, and Hootie chat built into the nest. PromptOwl agents run on top.
- Governed and enterprise-ready
- Turn on review and a named owner approves before AI may use it, every version is kept, and every read is recorded. OIDC single sign-on with Entra ID, Okta or Google Workspace comes with Enterprise.
Proof
Turn on review and changes wait for an owner. Every read is on the record.
This is ContextNest, not a mockup: with review on, owners approve what changes and agents read the approved version, and you can see who read what.
ContextNest Community 1.28, fictional demo company.
Research preprint, not peer reviewed
Better accuracy. Lower cost.
- ⅓
- of the input tokens of keyword search (BM25)
- 97%
- answer pass rate vs. 90–93% for BM25, stale-version test
- Same
- context for the same question, every time
With co-authors from
Get the whitepaper →Compare the bill: what each stack costs to run →Build your business case →Pick your path
Start from where you are.
Agencies & Organizations
You bring AI to your clients
Agencies, consultancies and fractional practices. Build your method into AI once, under your brand, and deploy it to every client. Run your own firm on it first.
Communities & Associations
You run an association, community or course
Your members already come to you for what you know. Turn it into an assistant that answers from your approved material, under your brand.
Enterprise
You're accountable for AI across the company
Whether something just went wrong or you've been handed an AI-first mandate, govern one team first and scale when your requirements say so.
Agents
You build and run agents
One governed source every agent reads from: versioned, approved and model-agnostic. Point your existing agents at it without a rewrite.
Start with your team
Pick a team and start from its own knowledge.
Give every AI your team uses the same approved knowledge.
Build the governed context your team's AI should already have, with every change versioned and every read recorded from day one. Works with Claude, Cursor and any MCP client.
What is ContextNest?
ContextNest is governed context for your AI: your company's knowledge as structured, graph-linked Markdown, versioned and approved by the people who own it. Any MCP-compatible AI (Claude, Cursor, Antigravity) reads directly from it, so every tool works from the same approved version. The Desktop app is free and runs locally. Patent pending.
How is ContextNest different from RAG or a vector database?
ContextNest handles living knowledge: structured, versioned, human-approved documents your AI reads directly, so the same question pulls the same context every time. RAG handles static knowledge: retrieval over large corpora that rarely change. PromptOwl governs both. Our research preprint (not peer reviewed), with co-authors from IBM Research and Emory University, measured a 97% answer pass rate versus 90–93% for keyword search (BM25) in a stale-version test, at about a third of the input tokens.
Read the whitepaper →How much does it cost?
The ContextNest CLI, engine and MCP server are open source (AGPL-3.0) and free. Install the CLI and it runs locally against your Markdown vault. The team workspace is free for up to 10 users, on our cloud or your servers. Hosted is $10 per user per month on our cloud. Self-Hosted is $10 per user with a 30-user minimum, unlimited storage and no overage bills. Every paid plan starts with a free month. Enterprise is custom, on one annual contract. Your own AI tools reading a nest over MCP never use credits. Credits only apply when PromptOwl supplies the model.
Can our compliance team approve this?
Every plan runs on the same SOC 2 and HIPAA controls. SOC 2 certification and reporting and a HIPAA BAA come with Enterprise. Role-based access, approval before AI may use a document, and a record of every read are part of the product.