Cross-tool work threads
Email, chat, docs, tasks and CRM from one team, joined by person and project.
INCLUDESThread IDs, stable pseudonyms, timestamps, outcomes
JSONL · ParquetRequest a sample →
Licensed, de-identified data from 100+ companies. For pre-training, SFT, RL and evals. Work data, long-form video, screen recordings and audio, linked in time order.
The same company data, shaped for the stage you are working on.
Years of real business writing, code and calls. Tokens the open web never had.
tokens · documents · threadsHow professionals really wrote the reply, fixed the bug or closed the ticket. With what happened next.
prompt · response · outcomeReplayable workspaces built from real company records, with tasks you can check.
state · action · rewardPrivate tasks that were never online, so scores stay free of contamination.
task · rubric · answer keyEvery dataset is built from licensed company data, de-identified and documented. Samples are available under NDA.
Email, chat, docs, tasks and CRM from one team, joined by person and project.
Tickets and live chats from first message to resolution.
CRM stages, emails and call notes from first touch to close.
Repositories, pull requests, reviews, linked tickets and incidents.
Meeting recordings and transcripts, with the follow-ups they triggered.
ERP orders, invoices and approvals, with the emails around them.
Head and chest camera footage of people doing hands-on tasks at work.
Multi-camera footage of real operations, shifts and incidents.
Hours of continuous footage of real jobs, segmented into steps with start and end times.
People completing real tasks in real software, recorded with every action.
Channel-separated customer calls in many languages, with transcripts.
Teleoperation sessions with joint states, actions and camera streams.
CAD and BIM revisions, site scans and the project email around them.
Tell us the modality, industry and volume. We source it from licensed archives, or pay companies to capture it.
Volumes vary by dataset. We share a manifest with record counts, date ranges and audit results before you commit.
Public text shows what people publish. Our data shows how teams plan, argue, decide and deliver over months.
Never published or crawled.
One ID per person, in every tool.
Projects, teams and tools as they change.
Typos, reversals and side chats left in.
Video, audio, screen and 3D, captured in real workplaces. Synced, de-identified and labeled to your spec.
Most video datasets are short clips. Real work takes hours. We deliver full recordings, with every step marked in time.
When did the technician replace the filter, and what did they check first?
01:02:24 to 01:15:36. The pressure gauge, at 00:41:10.
Every record keeps its stage history, so you can trace any line back to its license.
Before you commit, you get a manifest: sources, counts, dates and audit results.
| Source | Objects | Linked by |
|---|---|---|
| Messages, threads, calendar events | thread_id, person_id | |
| docs | Docs, sheets, decks, wiki pages, edit history | doc_id, person_id |
| chat | Channels, threads, replies, shared files | thread_id, person_id |
| meetings | Transcripts, notes, attendee lists | meeting_id, person_id |
| crm | Accounts, deals, stage history, activity | account_id, deal_id |
| support | Tickets, conversations, resolutions | ticket_id, account_id |
| tasks | Issues, sprints, status changes | issue_id, project_id |
| erp | Orders, inventory, invoices, approvals | order_id, account_id |
| code | Commits, pull requests, reviews | repo_id, pr_id |
| design | CAD and BIM drawings, design files, revisions | project_id |
| media | Video, audio and images, with transcripts and metadata | asset_id, person_id |
| robotics | Robot logs and teleoperation streams | device_id, asset_id |
All IDs are pseudonymous and consistent within a company.
{
"company": "co_0217",
"source": "chat",
"thread": "th_88f1",
"ts": "2024-03-11T09:42:10Z",
"author": "PERSON_14",
"text": "[PERSON_22] can you push the fix
for [CLIENT_3] before the demo?",
"links": [
{"source": "code", "pr": "pr_5531"},
{"source": "tasks", "issue": "iss_1190"}
],
"redactions": ["person", "client"]
}
Sources this company used. 140 people, 6 years. Log scale.
Same company. Tools were adopted at different times.
We replace identifiers with stable codes, so you lose the names and keep who talked to whom, about what, and when.
From: priya.n@northwind.io To: john@acme.com "Hi John, the Acme renewal is attached. Call me on +91 98200 41177 if the 12% discount needs sign-off from Rahul."
From: PERSON_14 To: CLIENT_PERSON_203 (CLIENT_3) "Hi CLIENT_PERSON_203, the CLIENT_3 renewal is attached. Call me on PHONE if the 12% discount needs sign-off from PERSON_22."
One contact, one contract. Scoped to what you are training.
Tell us the sources, industries and modalities you need. We share samples and a manifest under NDA.
Samples · manifest · audit logWe agree volumes, date ranges and labels, then sign one license covering every company in the order.
One license · consent on recordDelivered to your cloud bucket in documented formats, with release notes and PII audit results.
JSONL · Parquet · MP4 · MCAPFrom companies that own it and have signed a license with us. They range from 20-person startups to enterprises, across software, services, manufacturing, finance and more. Some are running, some have closed.
We check company identity and ownership, review each tool's terms and the seller's client contracts, and get a signed license before any export. Data that cannot be shared is excluded at source.
Automated detection, then human review. Identifiers become stable pseudonyms so cross-tool structure survives. Each release ships with sampled audit results, and you sign a no re-identification clause.
Yes. Tell us the industries, company sizes, regions, tools and date ranges you need. We source against that brief.
JSONL or Parquet with a documented schema, delivered to your cloud bucket. Raw files such as PDFs, drawings and recordings come with metadata in the same schema.
Yes. We share one de-identified company under NDA, with its manifest and audit log.
Yes. We license existing footage and logs from our sellers, and can run paid capture with their teams for first-person video, teleoperation and site recordings.
Under NDA, with its manifest and audit log.
Licensed, de-identified data from 100+ companies. For pre-training, SFT, RL and evals. Work data, long-form video, screen recordings and audio, linked in time order.
The same company data, shaped for the stage you are working on.
Years of real business writing, code and calls. Tokens the open web never had.
tokens · documents · threadsHow professionals really wrote the reply, fixed the bug or closed the ticket. With what happened next.
prompt · response · outcomeReplayable workspaces built from real company records, with tasks you can check.
state · action · rewardPrivate tasks that were never online, so scores stay free of contamination.
task · rubric · answer keyEvery dataset is built from licensed company data, de-identified and documented. Samples are available under NDA.
{{d.desc}}
Tell us the modality, industry and volume. We source it from licensed archives, or pay companies to capture it.
Volumes vary by dataset. We share a manifest with record counts, date ranges and audit results before you commit.
Public text shows what people publish. Our data shows how teams plan, argue, decide and deliver over months.
Never published or crawled.
One ID per person, in every tool.
Projects, teams and tools as they change.
Typos, reversals and side chats left in.
Video, audio, screen and 3D, captured in real workplaces. Synced, de-identified and labeled to your spec.
Most video datasets are short clips. Real work takes hours. We deliver full recordings, with every step marked in time.
When did the technician replace the filter, and what did they check first?
01:02:24 to 01:15:36. The pressure gauge, at 00:41:10.
Every record keeps its stage history, so you can trace any line back to its license.
Before you commit, you get a manifest: sources, counts, dates and audit results.
| Source | Objects | Linked by |
|---|---|---|
| Messages, threads, calendar events | thread_id, person_id | |
| docs | Docs, sheets, decks, wiki pages, edit history | doc_id, person_id |
| chat | Channels, threads, replies, shared files | thread_id, person_id |
| meetings | Transcripts, notes, attendee lists | meeting_id, person_id |
| crm | Accounts, deals, stage history, activity | account_id, deal_id |
| support | Tickets, conversations, resolutions | ticket_id, account_id |
| tasks | Issues, sprints, status changes | issue_id, project_id |
| erp | Orders, inventory, invoices, approvals | order_id, account_id |
| code | Commits, pull requests, reviews | repo_id, pr_id |
| design | CAD and BIM drawings, design files, revisions | project_id |
| media | Video, audio and images, with transcripts and metadata | asset_id, person_id |
| robotics | Robot logs and teleoperation streams | device_id, asset_id |
All IDs are pseudonymous and consistent within a company.
{
"company": "co_0217",
"source": "chat",
"thread": "th_88f1",
"ts": "2024-03-11T09:42:10Z",
"author": "PERSON_14",
"text": "[PERSON_22] can you push the fix
for [CLIENT_3] before the demo?",
"links": [
{"source": "code", "pr": "pr_5531"},
{"source": "tasks", "issue": "iss_1190"}
],
"redactions": ["person", "client"]
}
Sources this company used. 140 people, 6 years. Log scale.
Same company. Tools were adopted at different times.
We replace identifiers with stable codes, so you lose the names and keep who talked to whom, about what, and when.
From: priya.n@northwind.io To: john@acme.com "Hi John, the Acme renewal is attached. Call me on +91 98200 41177 if the 12% discount needs sign-off from Rahul."
From: PERSON_14 To: CLIENT_PERSON_203 (CLIENT_3) "Hi CLIENT_PERSON_203, the CLIENT_3 renewal is attached. Call me on PHONE if the 12% discount needs sign-off from PERSON_22."
One contact, one contract. Scoped to what you are training.
Tell us the sources, industries and modalities you need. We share samples and a manifest under NDA.
Samples · manifest · audit logWe agree volumes, date ranges and labels, then sign one license covering every company in the order.
One license · consent on recordDelivered to your cloud bucket in documented formats, with release notes and PII audit results.
JSONL · Parquet · MP4 · MCAPFrom companies that own it and have signed a license with us. They range from 20-person startups to enterprises, across software, services, manufacturing, finance and more. Some are running, some have closed.
We check company identity and ownership, review each tool's terms and the seller's client contracts, and get a signed license before any export. Data that cannot be shared is excluded at source.
Automated detection, then human review. Identifiers become stable pseudonyms so cross-tool structure survives. Each release ships with sampled audit results, and you sign a no re-identification clause.
Yes. Tell us the industries, company sizes, regions, tools and date ranges you need. We source against that brief.
JSONL or Parquet with a documented schema, delivered to your cloud bucket. Raw files such as PDFs, drawings and recordings come with metadata in the same schema.
Yes. We share one de-identified company under NDA, with its manifest and audit log.
Yes. We license existing footage and logs from our sellers, and can run paid capture with their teams for first-person video, teleoperation and site recordings.
Under NDA, with its manifest and audit log.