AI agents for Weights & Biases
Experiment tracking your reports can read. AGNT connects to Weights & Biases so agents can read what happens there, reason about it, and act — with a receipt for every run.
Weights & Biases and AGNT, in brief
| What it does | Lets an AGNT agent read from and act in Weights & Biases. |
|---|---|
| Connects with | Basic — held in the local vault |
| AGNT node | custom-api, or the zapier-action bridge |
| Runs when | A run completes |
| Typical build time | Ten minutes |
| Where data goes | Nowhere by default — AGNT runs on your machine and credentials never leave it. |
What you can automate
Reacts when…
- a run completes
- a metric crosses a threshold
Agents can…
- read runs and metrics
- summarise experiments
Teams use it for
- training runs summarised for the team
- regressions flagged with context
- experiment digests on schedule
Copy this Weights & Biases workflow
Design goal for Weights & Biases: training runs summarised for the team; regressions flagged with context; experiment digests on schedule. The template below supplies configurable nodes, not the complete logic for those goals.
Use this Weights & Biases setup template to arrange the schedule, input review and API operation. Configure the account-specific fields and add your own decision and approval controls before activation.
This is a schema-checked starting point, not an end-to-end tested automation. Replace every CONFIGURE_ value, choose the operations you intend, connect credentials and map source outputs into later inputs. An edge orders steps; it does not supply a condition, an approval gate or a field mapping. Add and test those controls explicitly before any write. Cloud APIs and model providers may receive data. Keep the workflow inactive until tested with fixtures.
{
"id": "63f1792a-e0d2-4784-a141-2dd2b4640738",
"name": "Weights & Biases — read, decide, act",
"nodes": [
{
"id": "c7f38ebd-ffd7-4165-a04e-b5ae48ae58d6",
"text": "Run on a schedule",
"x": 512,
"y": 144,
"isEditing": false,
"type": "trigger-timer",
"icon": "clock",
"category": "trigger",
"isSelected": false,
"parameters": {
"fireOnStart": "No",
"scheduleType": "Interval",
"schedule": "Every 15 Minutes"
},
"description": "Setup template for Timer Trigger. This trigger node fires the workflow at specified intervals or at a specific time. Configure credentials, inputs and output mapping before use.",
"error": null,
"isActive": false,
"output": null,
"outputs": {}
},
{
"id": "55ca0eb5-9bdd-4ae0-a9e2-00bb579b64c8",
"text": "Review source and draft",
"x": 512,
"y": 336,
"isEditing": false,
"type": "agnt-agent",
"icon": "agnt",
"category": "action",
"isSelected": false,
"parameters": {
"agentId": "CONFIGURE_AGENTID",
"message": "Decide what needs to happen in Weights & Biases. Explain your reasoning.\nSource input: CONFIGURE_SOURCE_INPUT. Produce a reviewable draft only; do not take external actions."
},
"description": "Setup template for Agent Chat. Chat with an AI agent from your agent library. Select an agent and send messages to interact with it within your workflow. Configure credentials, inputs and output mapping before use.",
"error": null,
"isActive": false,
"output": null,
"outputs": {}
},
{
"id": "e0570868-a33d-4ef9-a1e4-917d700e4302",
"text": "Call the Weights & Biases API",
"x": 512,
"y": 528,
"isEditing": false,
"type": "custom-api",
"icon": "connect",
"category": "action",
"isSelected": false,
"parameters": {
"url": "https://api.wandb.ai/graphql",
"method": "GET",
"authType": "None",
"headers": "{}",
"query": ""
},
"description": "Setup template for Custom API Request. This action node makes a custom API request to any endpoint with configurable method, headers, and authentication. Configure credentials, inputs and output mapping before use.",
"error": null,
"isActive": false,
"output": null,
"outputs": {}
}
],
"edges": [
{
"id": "6c52c7ff-4875-4a4e-aa74-07196073c6ec",
"start": {
"id": "c7f38ebd-ffd7-4165-a04e-b5ae48ae58d6",
"type": "output"
},
"end": {
"id": "55ca0eb5-9bdd-4ae0-a9e2-00bb579b64c8",
"type": "input"
},
"startX": 800,
"startY": 168,
"endX": 512,
"endY": 360
},
{
"id": "d7c1994a-207e-4335-a658-cc1b1b098e86",
"start": {
"id": "55ca0eb5-9bdd-4ae0-a9e2-00bb579b64c8",
"type": "output"
},
"end": {
"id": "e0570868-a33d-4ef9-a1e4-917d700e4302",
"type": "input"
},
"startX": 800,
"startY": 360,
"endX": 512,
"endY": 552
}
],
"zoomLevel": 1,
"canvasOffsetX": 0,
"canvasOffsetY": 0,
"isTinyNodeMode": false
}Configure Custom API Request
This action node makes a custom API request to any endpoint with configurable method, headers, and authentication.
- url — The URL of the API endpoint Current example:
https://api.wandb.ai/graphql. - method — The HTTP method for the request Current example:
GET. - authType — The type of authentication Current example:
None. - query — Query parameters for the request (e.g., 'key1=value1&key2=value2') Current example:
. - headers — Optional headers for the request. e.g., {"Content-Type": "application/json"} Current example:
{}.
Run this against your own Weights & Biases
AGNT Community Core is free and runs on your own machine. Nothing about this workflow leaves it.
Experiment tracking that reports itself
Weights & Biases holds the record of what was trained and how it performed. Turning that into a written summary for people who will not open the dashboard is exactly the translation work an agent is suited to.
Regressions with context
A metric crossing a threshold is more useful when the notification also carries what changed between runs. That comparison is available in the data and rarely assembled by hand.
How to reach Weights & Biases from a workflow
There is no first-party Weights & Biases node yet, and a page implying otherwise would cost you an afternoon. What works today is the generic custom-api node pointed at Weights & Biases’s own API, authenticating with basic.
| What you need | Value |
|---|---|
| Endpoint | https://api.wandb.ai/graphql |
| Method | POST |
| Authentication | Basic |
| Header | Authorization: Basic base64(api:{apiKey}) |
| Reference | Weights & Biases API documentation |
Worth knowing: The Basic username is the literal string `api`; the key is the password half.
The part that catches people out
The Basic username is the literal string `api`; the key is the password half.
Everything else about Weights & Biases behaves the way you would expect, so once that is handled the rest of the build is ordinary. The POST above is the shape every other call takes.
Connect Weights & Biases in two minutes
- Download AGNT Community Core — free, local-first, no account needed to run.
- Connect by pasting an API key into AGNT’s vault — stored encrypted on your machine, never uploaded.
- Drop the Weights & Biases node into a workflow or hand it to an agent — the first run produces a receipt you can read line by line.
Weights & Biases + AGNT — common questions
Can training runs be summarised?
Yes — read from the tracked runs and written for a non-specialist audience.
Can a metric threshold trigger a run?
Yes, including flagging regressions with the surrounding context.
How is it connected?
With an API key held encrypted in the local vault.
