Skip to content
Graphnosis Graphnosis

Auto-ingest from Your Tools

The first version of Graphnosis was a memory you talk to. The point of connectors is to make it a memory that grows on its own from the tools you already use. Six built-in connectors pull or receive new content on a schedule, ingest it into the engram of your choice, and let your AI clients recall it the same way they recall anything else you’ve added.

How connectors work

Every connector follows the same pattern:

  1. You install it once in Settings → Connectors in the Graphnosis app.
  2. You choose a target engram for ingested events.
  3. You paste credentials (or accept an auto-generated webhook URL).
  4. Graphnosis pulls on a schedule (default: every 15 minutes) — or webhook-style connectors fire on push.
  5. Each new event becomes a memory node in your engram, available to every AI client connected to your Cortex.

Why BYO credentials. Most cloud memory products use a “first-party OAuth app” model — you log into Slack via their app, they hold a token on your behalf, they’re a 4th-party data processor in every workflow. Graphnosis goes the other way: you create your own app/key in each service’s developer console and paste credentials into Graphnosis. The trade is 1–5 minutes of upfront setup-in-service per connector, and the win is end-to-end privacy — Graphnosis is never in the OAuth callback chain, the service-side relationship is between you and that service.

Credentials are encrypted at rest. As of v0.6.1, every connector credential (PATs, tokens, API keys) is XChaCha20-Poly1305 encrypted with your Cortex data key before it touches disk. Same crypto primitive as your .gai memory files. You can safely sync your Cortex folder via iCloud Drive / Dropbox / S3 — cloud providers see ciphertext only.

RSS / Atom feeds — the simplest connector

What it pulls: new entries from any RSS or Atom feed URL(s) you provide. Deduplicates by entry guid (or link as fallback), so re-pulls of already-seen entries are no-ops.

Setup time: 30 seconds. No credentials.

How to add

  1. Open Settings → Connectors → RSS
  2. Paste one feed URL per line. Examples:
    • https://news.ycombinator.com/rss
    • https://simonwillison.net/atom/everything/
    • https://anthropic.com/news/rss.xml
  3. Pick the target engram
  4. Click Save

The connector starts pulling on the next 15-minute tick. Click Pull now on the row to fetch immediately.

Tips

  • Quality > quantity. A handful of high-signal feeds (industry leaders’ blogs, primary-source news) is better than 50 noisy ones. Your engram will reflect what you feed it.
  • Multi-engram strategy. Use separate engrams per topic: a tech-news engram for HN/Simon Willison, a industry-x engram for the trade publications you actually read, etc.

GitHub — repos you care about

What it pulls: issues, pull requests, and releases from a list of repos. Configurable per-event-type so you can pull only releases for one repo and full issue/PR history for another.

Setup time: ~3 minutes.

Step 1: create a Personal Access Token

  1. Go to github.com/settings/tokens?type=beta (Fine-grained tokens — recommended over Classic PATs)
  2. Click Generate new token
  3. Token name: “Graphnosis” or similar
  4. Expiration: 90 days or longer
  5. Repository access: select the specific repos you want indexed (don’t grant all-repos access — least privilege)
  6. Permissions: under “Repository permissions” → Contents: Read, Issues: Read, Pull requests: Read, Metadata: Read (the last one is required)
  7. Click Generate token and copy it (github_pat_...)

Step 2: add the connector in Graphnosis

  1. Settings → Connectors → GitHub
  2. Paste the PAT
  3. Repos to watch: comma-separated owner/repo format. Example: anthropics/claude-code, nehloo-interactive/graphnosis-app
  4. Event types: check the boxes for what you want indexed (Issues, PRs, Releases)
  5. Pick the target engram
  6. Click Connect

Tips

  • Releases-only is great for “stay current.” Check just the Releases box and add a list of dependencies you care about — your engram becomes a personalized changelog.
  • Issues/PRs ingest is heavy. A busy repo with hundreds of open issues will produce hundreds of memory nodes on first pull. Start with one repo, see how it lands in your engram, then add more.
  • Token rotation: fine-grained tokens have an expiration. When yours expires, GitHub stops responding; the connector shows “auth expired” in Settings. Re-generate and re-paste.

Slack — your workspace as memory

What it pulls: starred items (default) and channel history (optional) from a single Slack workspace.

Setup time: ~5 minutes. The most involved of the connectors because Slack requires a custom app.

Step 1: create a Slack app

  1. Go to api.slack.com/apps
  2. Click Create New App → From scratch
  3. App name: “Graphnosis” (or anything)
  4. Workspace: pick the workspace you want to ingest from
  5. Click Create App

Step 2: configure scopes

  1. In the new app’s sidebar, click OAuth & Permissions
  2. Scroll to Scopes → Bot Token Scopes (or User Token Scopes if you want to pull your personal stars / personal DMs)
  3. Add scopes based on what you want pulled:
    • stars:read — required for starred items
    • channels:history + channels:read — required for public channel history
    • groups:history + groups:read — required for private channel history (only if invited to the channels)
    • users:read — recommended for human-readable usernames in ingested messages

Step 3: install the app to your workspace

  1. Scroll to the top of OAuth & Permissions
  2. Click Install to Workspace
  3. Approve the permissions
  4. Copy the Bot User OAuth Token (xoxb-...) — or the User OAuth Token (xoxp-...) if you used User scopes

Step 4: add the connector in Graphnosis

  1. Settings → Connectors → Slack
  2. Paste the token
  3. Check the boxes for what to pull (Starred items, Channel history)
  4. Pick the target engram
  5. Click Save

Tips

  • Starred items is the high-signal path. A workspace-wide channel-history pull is enormous and mostly noise. Star the messages you actually want indexed and let the connector do the rest.
  • User vs Bot tokens: User tokens pull what you see (DMs, your starred items, private channels you’re in). Bot tokens pull what the bot sees (must be added to channels). Pick based on what you want indexed.
  • Don’t grant chat:write scopes unless you want the bot to be able to post on your behalf. Read-only is sufficient for ingestion.

Trello — boards and cards

What it pulls: cards + checklists from a list of boards you specify, with their descriptions, labels, and checklist progress.

Setup time: ~3 minutes.

Step 1: create a Trello Power-Up to get an API key

  1. Go to trello.com/power-ups/admin
  2. Click New to create a Power-Up (you don’t need to publish it — it’s just to get an API key)
  3. Fill in the basic fields (any name + your workspace)
  4. After creation, go to the API Key tab
  5. Copy the API Key
  6. On the same page, click the Token link to generate a personal token (it’ll open Trello’s auth flow); copy the token

Step 2: find the board IDs you want to ingest

The board ID is in the URL when you have a board open:

https://trello.com/b/<BOARD_ID>/your-board-name
                       ^^^^^^^^^^

Copy that segment for each board you want indexed.

Step 3: add the connector in Graphnosis

  1. Settings → Connectors → Trello
  2. Paste API Key + Token
  3. Paste board IDs (comma-separated)
  4. Pick the target engram
  5. Click Save

Tips

  • One engram per board works well if boards represent distinct projects. Add the connector multiple times (one per board, with a different id slug each time) and route each to its own engram.
  • Archived cards are pulled in some Trello plans but not others. If you’re missing cards you’d expect, check whether they’re archived on the board.

Linear — issues and projects

What it pulls: issues from your Linear teams with their priority, state, assignee, and labels.

Setup time: ~1 minute. Easiest of the credentialed connectors — no OAuth flow, no app to create.

Step 1: create a personal API key

  1. Go to linear.app/settings/api
  2. Click Create new key
  3. Label: “Graphnosis” or similar
  4. Copy the key (lin_api_...)

That’s it. Linear’s personal API keys are first-class — no OAuth dance, no app review.

Step 2: add the connector in Graphnosis

  1. Settings → Connectors → Linear
  2. Paste the API key
  3. (Optional) team key (e.g. ENG, OPS) — leave blank to pull from every team you have access to
  4. Pick the target engram
  5. Click Save

Tips

  • Team filter is your friend. Without it, the connector pulls every issue from every team you’re a member of — for a large org this can be thousands of items on first pull. Specify a team key to narrow scope.
  • State + priority filters are coming in a follow-up release. For now, the connector pulls all states (Backlog, In Progress, Done, Cancelled, etc.).

Webhook — anything that can POST

What it does: generates a unique URL that any external service can POST events to. Push-only (no pull schedule).

Setup time: under a minute.

How to add

  1. Settings → Connectors → Webhook
  2. Pick the target engram
  3. Click Save
  4. The connector creates and shows a unique webhook URL like: 
    http://localhost:3458/webhook/my-webhook/a1b2c3d4-...-9876
  5. Click Copy and paste into your external tool’s webhook target

What to POST

Send a JSON body like:

{
  "text": "The full content to remember",
  "label": "Optional short title shown in the Sources list",
  "source": "Optional source attribution"
}

Only text is required. Anything you POST hits your local Graphnosis instance directly — no cloud relay.

What to wire it to

  • Zapier / IFTTT / Make: any “Webhook” action — sends new tweets, new Substack posts, new Discord messages, anything those platforms support.
  • GitHub Actions: push a curl step after a build to log build outcomes.
  • Cron + curl: scheduled shell scripts that summarize logs / metrics / whatever and POST the summary.
  • Browser bookmarklet: save the current article URL + selection. (See the docs section on bookmarklet for a one-line script — coming in v0.7.)
  • iOS Shortcuts: Action → “Get Contents of URL” → POST to the webhook URL. Lets you ingest from the Share Sheet on your phone.

Network reachability

The webhook URL is localhost:3458 by default — only reachable from the same Mac. For external services to reach it:

  • From your LAN: use the wizard’s mobile-setup flow to bind the MCP HTTP bridge to all interfaces. (Note: webhooks ride a separate port, 3458. Same security tradeoff.)
  • From the public internet: use a tunnel like Cloudflare Tunnel, ngrok, or Tailscale Funnel. The webhook token in the URL is your authentication.

Troubleshooting

“auth expired” status on a connector

  • GitHub PATs, Slack tokens, Trello tokens, and Linear keys all have expiration / rotation policies. When yours expires, the service returns 401 and the connector surfaces “auth expired.” Re-generate in the service, paste the new value via Edit on the connector row, save.

Connector shows “error” with cryptic message

  • The error message comes verbatim from the service’s API. Common cases:
    • GitHub: 404 usually means the token doesn’t have access to the repo (most often: you watch the repo but the PAT scoping doesn’t include it).
    • Slack: not_authed means the token is wrong; missing_scope means you didn’t grant the scope at app install (re-install the app with new scopes).
    • Trello: invalid token means the personal token was revoked; regenerate.
    • Linear: Authentication failed means the API key is wrong or rotated.

Pulls happen but events don’t show up in the engram

  • Check the target engram dropdown in the connector’s row — events go to the engram set at install time. If you’ve since renamed or archived that engram, the connector still references it but the engram isn’t in the active set.
  • Verify in Sources that the connector’s source.<id> entries are appearing. If yes but no nodes are showing in your atlas / recall, your engram might need a refresh (Settings → Cortex Tools → Refresh stats).

Too many events on first pull

  • Connectors don’t have a “first-pull limit” today. A busy GitHub repo or a starred-rich Slack workspace can drop hundreds of memory nodes the first time. If this happens, you can forget the source from the Sources list (one source covers all events from that connector since install) and re-install with tighter filters.

Security and storage notes

  • Credentials at rest: XChaCha20-Poly1305 encrypted with your Cortex data key in <cortex>/settings.json (as a credentialsEnc blob). Cloud-sync-safe — providers only ever see ciphertext.
  • In-memory credentials: decrypted on Cortex unlock; held in memory for the connector to use. Never logged.
  • No telemetry: connector calls go directly from your Mac to the service. Graphnosis has no server in the path.
  • Per-connector revocation: to revoke a connector’s access, Remove it in Settings (deletes the credentials locally) AND revoke the token in the service’s developer console (forces immediate cutoff even if the local copy somehow leaks).