How to Automate Fireflies Transcripts with the GraphQL API

November 16, 2026

The Fireflies API is GraphQL at https://api.fireflies.ai/graphql. You authenticate with a 32-character API key from Settings, and every query costs credits against your plan, so a scheduled job needs some care about volume.

This is for teams whose meeting data needs to live somewhere else — a CRM record, a support ticket, a weekly digest — not just in the Fireflies app.

How do you authenticate and test the connection?

Generate the key under Settings → Developer Settings → API Key. Store it as a bearer token in your environment.

Start with a small query in the API playground:

query { transcripts(limit: 3) { id title date duration } }

If you get an error about rate limits or permissions, check that the key belongs to a user on a plan with API access. Free plans have narrower limits and slower response windows, and the 404-vs-403 distinction is not always clear in the error message.

Rate limits are documented per plan and applied per minute. For any job that touches more than a handful of meetings, add a 1-second sleep between requests and handle 429 by backing off, not retrying immediately.

How do you pull a transcript with speakers and timestamps?

The transcripts query supports a transcript subfield returning sentences. Request more than id and title — the sentence-level data is what makes the transcript useful:

query($id: String!) {
  transcript(id: $id) {
    title date duration
    participants
    sentences { speaker_name text start_time end_time }
  }
}

Request only the fields you need. GraphQL lets you ask for everything, and asking for everything noticeably slows large responses.

For enumerating recent meetings, use the fromDate and toDate arguments on transcripts with ISO timestamps. Paginate with limit at 50 and skip, and stop when a page returns fewer rows than the limit — do not assume a total count field exists on all plans.

How do you generate a summary through the API?

Fireflies has a summary generation mutation. It returns a structure rather than one string, typically covering an overview, action items, and per-topic notes.

Call it on transcripts that lack a summary rather than regenerating everything — repeat generation spends credits and can produce different text for the same meeting, which is confusing when someone compares two versions of a client call.

If you need a custom summary format, generate it yourself: pull the sentences, concatenate with speaker labels, and send to a model you control. That gives you a stable structure and a cost you can predict. The trade-off is that you now own the summarisation quality.

How do you sync meetings into a CRM?

The pattern that works:

  1. Schedule a daily job that lists transcripts from the last 48 hours.
  2. Filter to meetings whose participants include an email on your customer domain.
  3. Match the external participant’s email to a CRM contact.
  4. Write the summary and a link to the Fireflies recording onto a note or activity record.
  5. Log the transcript ID in your own store so the job is idempotent and does not write duplicates.

Point 5 is not optional. Without a dedupe table, a re-run after a partial failure writes the same note twice and the CRM record looks careless to whoever reads it.

Keep summaries out of CRM fields that sync to email tools. Nobody wants an auto-generated call summary landing in a nurture sequence.

What are the real constraints?

Credits. Transcription and storage are metered, and API-generated summaries consume them too. A 500-meeting-per-month team on a mid-tier plan will hit limits faster than expected if the job regenerates summaries on every sync.

Speaker attribution remains imperfect. speaker_name comes from the join metadata, so a phone-only attendee often shows as an email address or a generic label.

And recordings expire depending on plan and settings. If your company has a retention policy that deletes transcripts after 90 days, your job must copy anything it wants to keep. Do not treat Fireflies as the archive.

Quick Summary

  • Fireflies exposes a GraphQL API at api.fireflies.ai/graphql, authenticated by a 32-character key.
  • Request sentence-level data with speaker labels; it is the part that makes transcripts useful.
  • Paginate with limit and skip, and back off on 429 rather than retrying hard.
  • Make CRM sync idempotent with a local dedupe table keyed by transcript ID.
  • Plan for retention: recordings expire, so copy what you need to keep.