JUMP TO

GET STARTED

Quickstart

Give Claude Code organisational memory in about five minutes.

1
Create an account and an API keySign up, then open API Keys → Create Key. Copy the key — it's shown once. Keys start with nv_live_.
2
Install the CLI
bash
```
npm install -g @nova-oneda/cli
```
3
Connect Claude Code
bash
```
nova-oneda setup --key nv_live_...
```
4
Save your first memory, then recall itIn Claude Code, say:

“Remember that we chose Postgres for billing, in Nova Oneda”

“Search my Nova memory for billing decisions”

Open a new session and ask about billing — Claude pulls the decision in automatically. You'll also see it on your Timeline.

TIPAlready have Jira, Confluence or GitHub connected in Claude Code? Try “Import my project into Nova Oneda” — then see Imports for reliable phrasing.

Core concepts

Oneda stores memories: small, structured records of what your team decided and why. Memories belong to a workspace and are organised into topics, which can branch into sub-topics (for example Billing › Quota limits).

Decision: What was chosen, with its rationale.
ADR: An architecture decision record — context, decision, consequences.
Assumption: Something the code relies on being true. Worth flagging when it changes.
Note / Reference: Useful context or a pointer to where something lives.
Artifact: A file or document captured alongside the reasoning.
Edit: A correction to an earlier memory. History is kept.

How recall works

Recall is automatic: before substantial planning or implementation, Claude calls nova_context to pull relevant memories. You can also search explicitly with nova_search. You don't type tool names — Claude picks the tool from what you ask.

INSTALL

Installing the MCP server

Oneda's memory connects to your AI tools over MCP. There are two supported paths today, and one that isn't ready yet.

OPTION 1 · RECOMMENDEDClaude Code (CLI) OPTION 2 · BETA AUTHClaude.ai / Desktop / Mobile NOT SUPPORTED YETChatGPT / GitHub Copilot

Option 1: Claude Code (CLI) — recommended, most reliable

1. Install the CLI and 2. connect it to your workspace. Get your key from the API Keys page in the Oneda dashboard.

bash

npm install -g @nova-oneda/cli
nova-oneda setup --key nv_live_...

3. Start using it. Nothing else to configure — Claude Code picks up memory automatically. If you already have Jira, Confluence or GitHub connected in Claude Code, pull that context in directly by saying: “Import my project into Nova Oneda.”

Option 2: Claude.ai / Claude Desktop / Claude Mobile

Set this up once on Claude.ai or Desktop — a keyboard makes entering the URL and token easier. It syncs to Claude for iOS and Android automatically afterward. You can't add a new connector directly from the mobile app itself.

Go to Customize → Connectors → Add custom connector. On Team/Enterprise, an Owner adds it once under Organization settings → Connectors; members then connect from Customize → Connectors.
Server URL:
https://oneda.freeelatech.com/mcp
Open Advanced settings → Request headers and add a header named authorization — include the word Bearer and the space:
Bearer nv_live_...

BETAHeader-based auth is a beta Claude feature and can be inconsistent. If the connection doesn't work reliably, the CLI setup is the more dependable path today.

ChatGPT / GitHub Copilot

NOT SUPPORTED YETChatGPT's connector settings currently only accept OAuth client credentials — not a static API key or Bearer token, which is the only auth method Oneda's MCP server supports right now. GitHub Copilot has the same underlying gap. Both need OAuth support added on Oneda's side; it isn't built yet.

Custom agents (MCP)

Any MCP client that supports remote servers with a static auth header can connect.

Server URL: https://oneda.freeelatech.com/mcp
Header: authorization: Bearer nv_live_...
On connect: The server lists recent topics for the workspace, so a cold client knows memory exists.

Each key is scoped to one workspace. Give each agent its own key so you can see and revoke its access independently.

REFERENCE

Once connected, what to say

You don't call these tools directly — Claude decides when to use them based on what you ask.

YOU SAY	ONEDA DOES
{{ t.say }}	{{ t.does }}`{{ t.name }}`

TIPVague requests like “import my project” can silently do nothing if Claude doesn't pick the right tool. Naming the action (import / search / remember) and pasting the actual URL or key gets a reliable result every time.

Imports

Seed memory from what your team already wrote down.

GitHub: README, manifest, key docs and repo structure. Paste the full repo URL.
Jira: Recent issues from a project. Name the project key, e.g. ENG.
Confluence: Pages from a space. Name the space key.

RELIABLE PHRASINGVague requests like “import my project” can silently do nothing. Name the action and paste the actual URL or key: “Import https://github.com/owner/repo into Nova Oneda”.

WORKING WITH MEMORY

Writing memories that recall well

A good memory states the decision, the reason and the constraint in one or two sentences. Future agents match on meaning, so be specific.

WEAK“We use Postgres.”

STRONG“Billing uses PostgreSQL because ledger writes need multi-row transactions and the team already operates Postgres. Revisit if billing moves off the main cluster.”

Say why, not only what.
Name the constraint that made it true — that's what tells a future agent when it stops being true.
When a decision changes, record the correction rather than deleting the old memory.
If Claude proposes a new topic, it asks first. Decline and the memory is discarded, not stored.

Add a memory manually

From the dashboard's API Keys page, use Add a memory manually to seed decisions before Claude has captured anything.

Timeline & usage

Timeline shows every memory newest first. Filter by topic in the sidebar and by kind (Decisions, ADRs, Assumptions, Notes, Artifacts). Memory Usage shows memories created over the last 30 days, MCP requests, context calls, search calls and your plan.

To see the reasoning behind a memory, ask Claude to “Visualize the decisions behind our auth system” — nova_visualize renders the decision graph as ASCII.

ACCOUNT

API keys

Create keys from API Keys → Create Key. A new key is shown once — copy it immediately.
The keys table shows name, created, last used, status and uses per key.
Revoking a key ends its access immediately.
Quota used and calls this period are shown at the top of the page.

Your Anthropic API key

Oneda uses Claude to classify and organise what it remembers. Connect your own Anthropic key so that processing runs on your account and quota. The key is encrypted at rest and never shown again after saving. You can replace or remove it in Settings.

Don't have one? Create an Anthropic API key.

Deleting your account

Deleting your account revokes all API keys and deactivates your login immediately. Stored memories are kept for 30 days in case it was a mistake, then permanently deleted. To undo within that window, contact support — there's no self-service reactivation.

SECURITY

Security & governance

{{ g.t }}: {{ g.d }}

Team or Enterprise account? Sign in here, or talk to us.

Troubleshooting

{{ f.q }}: {{ f.a }}