xhigh tier.
Atlas keeps its own request and response names. It takes
lead_info and struct, returns a task_id, pages with limit and after, and reports completed where other /v1 routes report succeeded. A run is not a Job yet: poll it with GET /job-status/{task_id}, not GET /v1/jobs/{job_id}.Endpoints
All paths are underhttps://api.sixtyfour.ai/v1/atlas.
Authenticate with your
x-api-key header, as on every other endpoint.
Checking access
Atlas is enabled on request; talk to sales to turn it on.GET /access works for every organization. It returns whether yours has Atlas and the engines your API key’s team can run, default first, with each engine’s price.
Starting an investigation
POST /investigations returns 202 with the run’s task_id and the new investigation:
Other fields are ignored without an error. Atlas API or People Intelligence lists the fields only People Intelligence takes.
GET /job-status/{task_id}, or open workspace_url to watch it in the Atlas section of the Sixtyfour app. A completed run’s result has the fields of an Atlas job result, including structured_data and field_confidence when you send a struct. A run the agents decline also reports completed; see Declined runs.
Continuing an investigation
POST /investigations/{investigation_id}/runs starts another run on the same workspace and graph. It takes the same body as starting one, without lead_info, name and private, and returns the same 202 response. Continuing needs editor access to the investigation.
POST /people-intelligence-async with investigation_id and exclusive_run: true.
The investigation’s status covers all of its runs. After you continue a finished investigation it still reads completed until the new run’s agents start, so follow the new run with GET /job-status/{task_id}.
Listing and reading investigations
GET /investigations returns investigations, newest first, the page limit and a next_cursor. An API key lists and opens its team’s investigations and the ones shared with that team.
Each investigation, also returned by
GET /investigations/{investigation_id}, looks like this:
shared is null when your team owns the investigation. Otherwise it names the team that shared it and your tier on it: a viewer can read and export, an editor can also run and edit.
Investigation status
Reading the graph and workspace
Findings, links and events
Findings are the graph’snodes and the items of GET .../findings. Empty fields are left out.
Links are the graph’s
edges and the items of GET .../links:
Events are the agents’ turn summaries:
With activity access, the event list also includes tool calls and their
payload: args, result and thinking. A large payload has a non-null payload.storage_ref instead of inline values; GET /investigations/{investigation_id}/events/{event_id}/payload returns it in full. The request takes the event’s id. storage_ref is opaque.
Images and attachments
Job results and image fields link attachments by permanenthttps://api.sixtyfour.ai/v1/media/... URLs that need no API key and can be embedded. For organizations with customer-managed encryption, media URLs need the x-api-key header, so fetch them from your server.
Following a run live
GET .../events/stream sends the agents’ turn summaries over Server-Sent Events as they are written, in order. Omit after to replay from the start.
The
sessions frame before end carries the run’s cost, or null if the cost has not been recorded yet; read cost_credits from GET /investigations/{investigation_id} later. Comment lines keep the connection open; ignore them.
status and end follow the investigation’s status. After you continue a finished investigation, open the stream once it reads running.
A stream closes after about five minutes. Reconnect with the last event’s id as after or as a Last-Event-ID header; after also accepts next_cursor from GET .../events. Every new stream sends status and sessions again, so treat them as the current state, not as changes.
Browser EventSource can’t send the x-api-key header, so read the stream with an HTTP client, as below. Close it on end or access_lost: the stream sets retry: 1000, and each reconnect counts toward your rate limit.
Reports
Atlas starts a report when a run completes. To get one at any other time, or to be sure it covers the newest findings:POST .../reportstarts a PDF report from the current workspace and returns 202 with a report build. A report build is not a Job: it has its own statuses, below. A request while a report is building joins that build; if the build predates newer findings, the job hasfollow_up: trueand a new build starts after it. If the latest report is current, the job iscompletedwithcached: true. Send{"retry_failed": true}to retry a failed build. An empty workspace returns 400.- Poll
GET .../reportuntilstatusiscompletedandfollow_upisfalse. It returns the latest job:job_id,status,follow_up,cached,requested_at,completed_at,errorandbyte_size. - Download
GET .../report.pdf.
If
follow_up stays true for 10 minutes after a build ends, send POST .../report again.
GET .../report.pdf answers:
When no run is going, you can also poll
report.pdf after POST .../report until it returns 200 with X-Report-Current: true. An older report that only lacks newer findings returns 200 with X-Report-Current: false.
Export
GET .../export.zip downloads everything visible in the app in one archive:
GET .../identifiers.csv lists every email address, phone number, IP address, name, organization, account and domain the investigation found, one row each, with its confidence, whom it is about, a source URL, the finding that names it and when it was first found. Retired findings are left out. An email or IP address read from a finding about something else leaves About and Relation empty. Cells a spreadsheet would run as formulas are escaped with a leading '; pass spreadsheet_safe=false to read the file as data.
Web app only
These have no API route and are done in the web app: stopping a run (Stop latest run), steering agents, attaching files to a run, sharing with another team or by public link, renaming and deleting investigations, folders, editing notes and findings, revising a report and reading earlier versions, automations, and saved instructions.Pricing
See Credit Costs for more information.Related
Org chart output
Start an investigation withtask: "org_chart", engine: "xhigh" and the company in lead_info. Use research_plan for an optional department or other scope. Org chart tasks use per-person billing; see Credit Costs.
Read GET /investigations/{investigation_id}/output for the latest saved structured output. The response contains data and confidence objects. Org chart data includes nodes, relationships and company_linkedin_id; nodes represent companies, teams and people. Output can be partial while research runs. Before the current run has saved output, both objects are empty. The endpoint uses the same investigation access rules as other reads.