Triage incoming RFQ emails. AI-extracts customer/area/qty/material/processes/due, files attachments to a per-RFQ Drive folder, appends to a Sheets quote log, auto-replies to the customer in thread with a reference number, and notifies the estimator. Processes one RFQ per scheduled run.

`rfq-triage`

Pulls one unprocessed RFQ email per scheduled run, files attachments to a per-RFQ Drive folder, has Claude extract customer/qty/material/processes/due/urgency from the message body, appends a row to a Sheets quote log, auto-replies to the customer in-thread with a reference number, notifies the estimator, and relabels the message.

Use case

A small machine shop receives RFQs by email. Without automation, someone has to read each one, save the drawings somewhere sensible, type the customer + qty + due date into the quote tracker, send an acknowledgment, and forward it to the estimator. This workflow does all of it. Schedule it every 5 minutes; backlog catches up automatically.

Setup

Credentials (Credentials view): - google-workspace — Google Workspace service-account JSON with domain-wide delegation - claude-api — Anthropic API key
Runtime environment (Runtime Environments view): - bash — InTouch auto-creates this on Linux/macOS install from /bin/bash. Must include bash, awk, grep, cut, tr, mktemp, date, python3, and the gog CLI on PATH. - Windows: InTouch auto-creates powershell and command prompt instead. To run this workflow on Windows: install Git Bash or WSL and add a bash runtime env in InTouch pointing at it (e.g. C:\Program Files\Git\bin\bash.exe), OR change runtimeEnvName: "bash" to "powershell" in workflow.yaml and rewrite the script blocks in PowerShell.
Gmail prep: - Create label RFQ - Create label processed - Create a filter that auto-applies RFQ to inbound customer RFQ emails (by sender domain, by quote request / RFQ in subject, or by forwarding from [email protected])
Drive prep: - Create a parent folder for RFQs (any name) - Copy its folder ID from the URL: https://drive.google.com/drive/folders/<THIS_IS_THE_ID>
Sheets prep: - Create a sheet with tab QuoteLog and 11 columns: ref_id | received_at | customer | from_email | project | qty | material | processes | due_date | urgency | status - Copy the spreadsheet ID from the URL
Edit workflow.yaml — replace the four placeholders: - <<RFQ_LABEL>> → RFQ - <<RFQ_PARENT_FOLDER_ID>> → the Drive folder ID - <<QUOTE_LOG_SHEET_ID>> → the Sheet ID - <<ESTIMATOR_EMAIL>> → the internal estimator's email - <<SHOP_NAME>> → company name for the auto-reply signature
Schedule the workflow every 5 minutes

Pipeline

intake — runtimeenv (bash + gog). Finds the oldest message labeled RFQ without processed, fetches it as --format full --plain TSV, parses headers + body + attachment rows, creates a dated subfolder under the RFQ parent, downloads each attachment and re-uploads it to Drive, emits a JSON blob.
extract — anthropic (claude-sonnet-4-6). Parses the JSON blob into structured fields: customer, project, qty, material, processes, due, urgency, complexity. Returns strict JSON.
dispatch — runtimeenv (bash + gog + python3). Appends a row to the Sheets quote log, sends an in-thread auto-reply to the customer with a RFQ-YYYYMMDD-HHMMSS reference, sends a separate digest to the estimator, then applies processed and removes RFQ on the original message.

If intake finds no new RFQ, the rest of the pipeline exits cleanly with no Sheets write, no emails, no API spend beyond one Anthropic call to confirm "empty."

What runs per scheduled fire

gmail messages search "label:RFQ -label:processed" --max 1
        ↓ message ID
gmail get --format full --plain
        ↓ headers + body + attachments
drive mkdir <date>_<subject> --parent <RFQs folder ID>
        ↓ subfolder ID
gmail attachment + drive upload (per attachment)
        ↓ JSON blob (message_id, thread_id, from, subject, body, folder, attachments[])
anthropic extract
        ↓ structured RFQ fields
sheets append QuoteLog!A:K <11 values>
gmail send --thread-id <thread> (customer reply)
gmail send (estimator notification)
gmail messages modify --add processed --remove RFQ

Customization

Schedule cadence — every 5 min for active shops, hourly for quieter ones; cost is one Anthropic call per fire regardless of whether there's work
Auto-reply wording — edit the --body of the customer-facing gmail send in the dispatch task
Estimator digest format — same task, second gmail send call
Sheet columns — if you change the QuoteLog column layout, update the sheets append value list and the column count in the range (QuoteLog!A:K → A:N etc.)
AI provider — swap anthropic for openai, gemini, or ollama task types; same prompt works
Reference ID format — currently RFQ-YYYYMMDD-HHMMSS; edit the REF_ID= line in dispatch
Multiple intake labels — change the search query in intake to OR multiple labels ("(label:RFQ OR label:quote) -label:processed")

Verified

The gog flag names (--parent, --out, --thread-id, --add/--remove, --reply-to-message-id) are all confirmed against the bundled gog v0.19.0 (2026-06-01). The gmail get --plain TSV shape and the drive upload/mkdir --plain output keys should be re-verified on first run against your Workspace account. If the bundled gog ever moves to a new major version, re-verify before promoting.

Source

See workflow.yaml.