One source of truth for your team, so every AI tool answers with the same current facts.
Verde stores your team’s knowledge once, then serves the current version to Claude, Cursor, ChatGPT, and any MCP client — so every tool answers from the same source, never last quarter’s.
- No AI API keys required
- Set up in minutes
- AI drafts, you publish
Speaks the open standard your tools already use
The problem
Your knowledge is scattered across every chat window.
Every AI tool your team uses keeps its own private memory. The pricing decision lives in one person’s Claude history, the architecture call in someone’s Cursor session, the support process in a doc nobody updates.
So your tools confidently repeat last quarter’s prices, new teammates re-ask answered questions, and nobody can say which version of a decision is the real one.
Verde replaces scattered chat histories with one permission-aware memory that every tool, and every person, reads from.
How it works
Write it once. Every tool remembers.
Three steps from scattered knowledge to one shared source of truth.
Capture it once
People and AI tools file structured memories (facts, decisions, policies, lessons) into one shared workspace instead of five chat histories.
Connect your tools
Generate a token, paste one MCP endpoint into Claude, Cursor, Codex, or ChatGPT. No plugins, no per-tool setup, no vendor lock-in.
Every answer is current
Retrieval always favors the current, published version. Update a decision once and every tool your team uses answers with the new version.
Every memory is structured markdown. Code, tables, and Mermaid diagrams render cleanly in both themes, with types, tags, sources, and dates attached.
See your AI think
Know what your AI knows — and where it comes up blank.
Every tool you connect leaves a trail. Verde turns it into a live picture of what your AI reads, leans on, and can't find — so your knowledge gets sharper the more it's used.
Knowledge gaps, found for you
Every search your AI runs that comes back empty is captured and ranked — each one a question your team never wrote down. One click turns it into a document.
See what your AI leans on
The documents pulled into AI context most, by real recalls and search hits. Know which knowledge is quietly load-bearing — and worth keeping sharp.
AI and team, side by side
Reads and writes per day, your AI tools next to your people. Proof the memory layer is being used — or a nudge that a tool isn't connected yet.
Knowledge gaps
Searches your AI ran that found nothing — each one a document worth writing.
- “refund window for annual plans”×6Write it
- “on-call escalation policy”×4Write it
- “SSO / SAML setup steps”×3Write it
Illustrative — your dashboard fills in from real AI activity.
Built-in privacy model
Three scopes. Zero accidental leaks.
Every memory lives in exactly one scope, and every rule is enforced on the server, never left to the AI client's good behavior.
Public
Anyone on the internet — no sign-in required
Your external knowledge base: product info, FAQs, docs, and public policies. Only published documents appear publicly.
Company
Signed-in members of the team
Internal knowledge: decisions, plans, processes, architecture, and lessons learned. Never visible outside the team.
Private
Only you — invisible to every other member
Your personal memory: preferences, working notes, drafts, and personal AI instructions. No other member — not even an owner — can browse it.
No owner or admin can browse another member’s private memories. The rule lives in SQL, and our end-to-end suite proves it on every release.
Why Verde
Memory you can actually trust.
Plenty of tools bolt memory onto AI. We built the layer underneath (governed, current, and shared) so the answers your tools give are ones you'd stand behind.
Governed, not a junk drawer
Draft → published → superseded. Knowledge has a lifecycle, an owner, and an audit trail, not an ever-growing pile of half-true notes.
History without confusion
Superseded knowledge stays queryable and clearly marked. You get institutional memory and a single current answer.
Security that doesn't trust the client
Every permission check runs server-side. A compromised or misbehaving AI client still can't read what its user couldn't.
Open standard, no lock-in
Standard MCP over HTTP. Switch AI vendors tomorrow and your memory layer comes with you, untouched.
Why not just…
A wiki, or your AI’s built-in memory?
Both leave a gap Verde is built to close.
A shared wiki or doc
People can read it — your AI tools can’t, not without wiring up each one. And nothing tells you when a page went stale, or that a decision it records was reversed months ago.
Built-in AI memory
Lives inside one tool, for one person, where no teammate or admin can see it. It can’t be reviewed, corrected, or trusted for anything your brand says out loud.
Verde
One governed source every MCP tool reads from. The current version always wins, a human gates anything public, and you can see exactly what your AI asked for.
Everything included
The details other memory tools skip.
AI drafts, humans publish
AI can draft knowledge but can never publish to your public KB on its own; a member with publish permission takes it live. Your brand never says something a model made up.
Current beats stale
Supersession links keep the full decision history, clearly marked, but the newest published version always wins retrieval. No more $19 answers after the price moved to $29.
Permissions enforced in SQL
Workspace, role, bucket, and private-ownership rules run server-side on every query, never trusted to the client. A private memory never crosses users, even for owners.
Search with zero AI keys
Weighted Postgres full-text search ranks published knowledge first, out of the box. No embedding pipeline to babysit, no per-query LLM bill.
Real documents, not blobs
Memories are structured markdown with types, tags, sources, and dates. Tables, code, and Mermaid diagrams render beautifully in both themes.
Tokens you can revoke
Each MCP token is shown once, stored only as a hash, and acts with exactly its owner's permissions. Revoke any connection in one click.
1
source of truth for people and AI
3
privacy scopes, enforced in SQL
10
MCP tools your AI can call
0
AI API keys required to search
Connect in minutes
One endpoint. Every tool your team runs.
Generate a token from the dashboard, paste one command, and your AI tools can search, draft, and update shared knowledge with exactly the permissions you’d have yourself.
claude mcp add --transport http verde \
https://getverde.ai/api/mcp \
--header "Authorization: Bearer YOUR_TOKEN"Works the same in Cursor, Codex, and any client that reads .mcp.json. OAuth 2.1 supported for Claude.ai connectors.
FAQ
Questions teams ask us.
What is MCP, and why does it matter?
The Model Context Protocol is the open standard AI tools use to reach external knowledge. Because Verde speaks standard MCP over HTTP, it works with Claude, Claude Code, Cursor, Codex, ChatGPT, and any future MCP-compatible client, so you're never locked into one vendor's plugin system.
Can an AI tool publish to our public knowledge base?
No. Anything an AI client creates for the Public bucket is saved as a draft and waits for a member with publish permission to take it live in the dashboard, even if the token belongs to a workspace owner. Public content is always human-published.
Can admins read my private memories?
No. Private memories belong to one member and are invisible to everyone else, including owners and admins. The rule is enforced in SQL on every query and verified by our end-to-end test suite.
Do we need an OpenAI or Anthropic API key?
No. Retrieval uses weighted Postgres full-text search, so there's no AI bill attached to search. The search service is isolated, so embeddings can be added later without a rewrite.
What happens when knowledge changes?
You supersede the old memory with the new one. History is preserved and clearly marked, but normal retrieval, whether human or AI, only returns the current version. Tools can explicitly ask for historical context when they need it.
How do AI connections authenticate?
Each member generates their own bearer token (shown once, stored as a SHA-256 hash). A token acts with exactly that member's permissions and can be revoked at any time. OAuth 2.1 is supported for clients like Claude.ai that prefer it.
Give your team and your AI one memory worth trusting.
Create a workspace, invite your team, and connect your first AI tool before your coffee gets cold.