Give an agent Chorus tools for transcripts and CRM write-back
This agent reads finished sales call recordings, pulls out the summary and action items, and writes them back to the CRM. It is a build guide for engineers, including the pitfalls that break a naive setup.
Paste this into ChatGPT, Claude or your agent tool. See the prompt
- Pipeline Hygiene / CRM
- advanced
- RevOps
- Category
- Pipeline Hygiene / CRM
- Difficulty
- Advanced
- Est. cost
- Not stated by source
Works with
Problem
Updating the CRM from meeting transcripts is manual. Wiring an LLM agent to the Chorus API naively breaks on async tasks, binary media and strict write-back payloads.
Setup
An engineering team binds Chorus API tools, generated as JSON schemas through a unified API layer, to an LLM agent. The agent lists recent conversations whose processing is complete, fetches each conversation with its summary, action items and linked deals, and pushes sales-qualification fields back to the CRM opportunity through the Chorus write-back endpoint.
Prompt
Starter prompt by The GTM Agent Weekly, written from the setup described on this page. It is not the source's own prompt; see the source link below for their write-up.
# ROLE
You process recent sales conversations from the call recorder and write sales-qualification
fields back to the matching CRM opportunity.
# TOOLS YOU HAVE
- list_conversations(since): recent calls and whether processing is complete.
- get_conversation(id): transcript summary, action items and linked deals.
- writeback(opportunity_id, fields): update qualification fields on the CRM opportunity.
# INPUTS
- {{SINCE}}: how far back to look. {{FIELDS}}: the qualification fields to fill, each with a
definition and allowed values (e.g. budget, authority, need, timeline, next step).
# TASK
1. List conversations since {{SINCE}}. Skip any whose processing is not complete; they will
be picked up next run.
2. For each, get the conversation. Skip it if it has no linked deal, and log "no deal linked".
3. Extract each field in {{FIELDS}} only where the summary or action items support it.
4. Write back only fields that changed. Never blank out an existing value.
# OUTPUT (per run)
A log line per conversation: id, opportunity, fields written, fields skipped and why.
# GUARDRAILS
- One write per opportunity per run. If two calls conflict, keep the most recent and log both.
- No values from guesswork. Missing is better than wrong.
- Do not process the same conversation twice; keep a record of processed ids. Replace every {{PLACEHOLDER}} before running. Works with any capable LLM and runner.
Tools
Steps
- Connect Chorus and fetch typed tool schemas for its endpoints.
- Have the agent list conversations filtered by date and a completed processing state.
- Fetch the full conversation object for summary, action items and linked CRM records.
- Fetch the correct CRM object IDs before calling the qualification write-back endpoint.
- Add rate-limit backoff and task polling to the orchestration layer.
Results
Integration guide; no outcome data.
What goes wrong
Gotchas
- Heavy operations return a task ID, not a result; an agent that treats the first OK as done reports success too early.
- Media endpoints stream the raw file by default; request JSON to get a download URL instead.
- The write-back rejects payloads without the right meeting, object type and opportunity IDs.
Get 5 like this every week, free.
Real agent setups with sources and failure modes, in one short weekly email.
Free. One email a week. Unsubscribe anytime.
Source and credit
Truto · vendor
Source dated 2026-10-11 · last reviewed 2026-10-11. Summarized in our own words; see the source for details.