LATEST NEWS
MEASUREMENTThe Fair Litmus: three task families, three passes - the cortex uplift, measured honestly2026-11-17 MEASUREMENTThe Self-Built Conscience: a 35B model constructed its own workshop and solved the task that defeated it ten times2026-11-17 MEASUREMENTThe Cortex Uplift Experiment: a 35B model, a corrupted database, and the wall we finally mapped2026-11-17 MEASUREMENTTerminal-bench round five: the first honest grading, the csv flip, and the openssl counter-flip — variance in one picture2026-11-15 MEASUREMENTTerminal-bench A/B, round four: the template fix worked (parroting 42% to 7%), the corpus didn't move resolution, and four rounds deliver the instrument's verdict2026-11-15 INCIDENTIncident disclosure: the experimenter broke the scaffolding too — our own failure-mode ledger from the three-round agent benchmark2026-11-15 MEASUREMENTTerminal-bench A/B, round three: the uplifted seat, a null result, and what three rounds actually measured2026-11-15 IMPROVEMENTThe vector question: what FaultLine's memory stores, and why2026-11-15 IMPROVEMENTThe convergence gate: corrections now land, every time2026-11-15 MEASUREMENTTerminal-bench A/B, round one: a cold seat, five honest empties, and a 33%→71% delta we cannot credit to memory2026-11-15 MEASUREMENTTerminal-bench A/B, round two: mini-swe-agent, a seeded seat, and memory that actually carried content2026-11-15 IMPROVEMENTDocuments are now a store of record: versioned, byte-returnable, superseded like facts2026-11-14 IMPROVEMENTRecall now answers plainly2026-11-14 IMPROVEMENTDocument ingest: snapshot-first storage — ingest can no longer clog serving2026-11-14 IMPROVEMENTServe latency: a stage-by-stage decomposition and a 38% reduction2026-11-14 INCIDENTR4-K: a pathological knowledge shape could pin a CPU worker indefinitely2026-11-14 IMPROVEMENTThe redaction fix: “remove this” now actually removes this2026-11-14 INCIDENTIncident disclosure: a heavy seat answered “I know nothing about your family” — while holding all of it2026-11-14 IMPROVEMENTThe heavy-seat walk fix: when 627 connections per question was the latency2026-11-14 DEPLOYMENTThe blended-family recall round: 53% to 97%, measured run by run2026-11-13 IMPROVEMENTThe recall overhaul: the numbers, the methodology, and the change log2026-11-12 MEASUREMENTThe Fair Litmus: three task families, three passes - the cortex uplift, measured honestly2026-11-17 MEASUREMENTThe Self-Built Conscience: a 35B model constructed its own workshop and solved the task that defeated it ten times2026-11-17 MEASUREMENTThe Cortex Uplift Experiment: a 35B model, a corrupted database, and the wall we finally mapped2026-11-17 MEASUREMENTTerminal-bench round five: the first honest grading, the csv flip, and the openssl counter-flip — variance in one picture2026-11-15 MEASUREMENTTerminal-bench A/B, round four: the template fix worked (parroting 42% to 7%), the corpus didn't move resolution, and four rounds deliver the instrument's verdict2026-11-15 INCIDENTIncident disclosure: the experimenter broke the scaffolding too — our own failure-mode ledger from the three-round agent benchmark2026-11-15 MEASUREMENTTerminal-bench A/B, round three: the uplifted seat, a null result, and what three rounds actually measured2026-11-15 IMPROVEMENTThe vector question: what FaultLine's memory stores, and why2026-11-15 IMPROVEMENTThe convergence gate: corrections now land, every time2026-11-15 MEASUREMENTTerminal-bench A/B, round one: a cold seat, five honest empties, and a 33%→71% delta we cannot credit to memory2026-11-15 MEASUREMENTTerminal-bench A/B, round two: mini-swe-agent, a seeded seat, and memory that actually carried content2026-11-15 IMPROVEMENTDocuments are now a store of record: versioned, byte-returnable, superseded like facts2026-11-14 IMPROVEMENTRecall now answers plainly2026-11-14 IMPROVEMENTDocument ingest: snapshot-first storage — ingest can no longer clog serving2026-11-14 IMPROVEMENTServe latency: a stage-by-stage decomposition and a 38% reduction2026-11-14 INCIDENTR4-K: a pathological knowledge shape could pin a CPU worker indefinitely2026-11-14 IMPROVEMENTThe redaction fix: “remove this” now actually removes this2026-11-14 INCIDENTIncident disclosure: a heavy seat answered “I know nothing about your family” — while holding all of it2026-11-14 IMPROVEMENTThe heavy-seat walk fix: when 627 connections per question was the latency2026-11-14 DEPLOYMENTThe blended-family recall round: 53% to 97%, measured run by run2026-11-13 IMPROVEMENTThe recall overhaul: the numbers, the methodology, and the change log2026-11-12
⊢FaultLine

Your AI forgets. FAULTLINE remembers.

Try it free
Try it free
A full seat, on the house while you decide.
  • ✓tell it things once — it remembers
  • ✓your own private space
  • ✓works with the AI you already use
⊢

Built, shipped, and running

Not a demo and not a promise — here is where things actually stand.

In production today

A live system with real connected users, and the full commercial layer already built — billing, subscriptions, seats, invoicing, terms.

Open and auditable

The core engine is public open source — you can read the architecture, not take our word for it. Hosted in Canada.

A real company behind it

Operated by Volenti, a Canadian software company that already ships and runs several products. FaultLine is one of them.

Read the honest inventory ▸