A CRM for product-led sales, fed by your product’s events in seconds
Three founders · Martín, Dewi and Florencia
Tringel is a CRM for sales teams whose customers sign up on their own. Its buyer is the head of sales at a product-led company of 20 to 150 people, where a ninth invite on an eight-seat plan can take hours to reach a rep.
Records in Tringel are written by the product being sold. Each usage event is pushed to us, matched to an account and checked against every live play, a three-line rule the head of sales writes. A play that fires puts the account and the reason on the rep’s list within a design budget of 60 seconds.
Tringel’s models match accounts, write the Why now brief and draft candidate plays. By our design estimate, exact rules match about seven workspaces in ten; an embedding model proposes matches for the rest. An open-weight 8B model writes each brief from 30 days of events, citing event ids; a sequence model drafts plays from what preceded past upgrades.
Only a play a person set live can put an account on a Tringel rep’s list. A match below 0.92 confidence waits for a person in the Unmatched queue; a candidate play stays a draft until the head of sales sets it live. Those decisions, reps’ brief edits and signal outcomes become labeled pairs that tune only that customer’s matching threshold, brief adapter and candidate plays.
Tringel’s GPUs wake for a new customer’s onboarding and for the morning a customer ships a release. Onboarding means matching and embedding 12 months of events, often tens of millions of rows, within hours; a release can fire hundreds of signals, each wanting a brief. Between bursts the GPUs scale to zero, while the intake, stream processor, database and web app run on CPU all day.
The usage events Tringel receives carry the names, emails and behavior of our customers’ own users. Those users were usually promised a fixed list of subprocessors, so inference runs on GPU capacity we control, beside the rest of the service on cloud infrastructure in Frankfurt. No event leaves the European Economic Area through us.
Tringel’s GPU plan moves in dated steps from December 2026 to June 2027. As of September 2026 the matching and brief models run on NVIDIA L4-class cloud GPU capacity we control in Frankfurt. Briefs get per-customer adapters by December 2026; the candidate-play model and backtests move to GPU by March 2027; account timelines get embedding retrieval by June 2027.
Tringel S.R.L. was founded in October 2023 and is registered in Córdoba, Argentina. The first two years went into the intake and account matching, because a signal on the wrong account is worse than none. We have just opened and have no results to publish yet.
The founders
Ran a six-rep team at an Amsterdam developer-tools startup and watched a Slack alert channel between calls. The signals he saw after lunch had happened before breakfast, and the CRM never said why he had called.
Martín EcheverríaFounder
Wrote the cron job that posted yesterday’s hot accounts to sales every morning, at two companies in a row. Both times the event had left the product within a second and then sat in a queue nobody owned.
Dewi PattinamaCTO
RevOps lead who spent four years asking reps in Slack where each account stood, because the record was a day behind. She wanted the reason for a call written down before the rep dialed.
Florencia BianchiCRO
Tringel S.R.L.
Boulevard San Juan 540, Piso 3, X5000ATT Córdoba, ArgentinaContact
tringel.pro