Brigitte
Overview
This is the daily workday summary of meetings. It ideally differentiates the protocol based on whether the meeting is with a VC or a startup. It ignores internal meetings and meetings with 10 or more unique company domains.
Access
Trigger
Scheduled daily at 6:30 AM. It finishes around 7:00 AM due to the intentional 60-second processing delays between team members.
The workflow
Brigitte workflow on the self-hosted n8n instance
Credentials
| name | place | type |
|---|---|---|
| OpenAIAPIKeyMilos.AlecGenerated | n8n | OpenAI credentials |
| PerplexityKeyAsBearerMilos | n8n | Bearer Auth |
| PerplexityAPIKeyMilos | n8n | Perplexity API |
| n8n-application-level | Entra/Enterprise | Application level |
| n8n-hv-application-level | n8n | OAuth2 API |
Technical owner
This is who is emailed in the case of runtime failures.
Common Tasks
Subscription management (manual)
There is an n8n-internal table keeping track of the IM team in the Brigitte subscription table. New entries or re-subscribing need to be managed manually.
Unsubscribing (automated)
To emphasize the contrast, unsubscribing is intended to need no manual management. Each user gets an individualized link at the bottom of each daily Brigitte email, which connects them to a webhook in this workflow and unsubscribes them with a success message.
Disable during holidays (manual, optional)
Xmas, Easter, etc.
Troubleshooting
Expired IAM credentials
The application-level credentials are set via Entra/Enterprise to expire after a period of time. Kilian has confirmed that the current ones should be good through September 2, 2026.
Perplexity failing
Currently the sole failure mode in this workflow is when one of the Perplexity API calls fails. This has occurred in the past because the credit card ran out or strict output formatting on the Perplexity side suddenly failed. An email is then sent to ping the technical owner.
- No 4xx differentiation: just a generic ping with limited execution info to milos.tadic@hvcapital.com
- Then inspect executions on n8n GUI for details
By contrast, an OpenAI API call will likely fail silently.
Ad infinitum
Technically, a recent patchy solution against the structured output failures is a “loop-until-success” and has no boundary limitation.