← Admin Panel

Harvester

admin panel · wireframes · data
shell loading

The data-collection hub: every source, its state and sync progress — the source model lives in Harvester.

Global collection settings
embedding model · schedule
Processing
Embedding model
inherited from the AI models registry — changed there too text-embedding-3-large ↗ 1
Incremental sync schedule
Default interval
data freshness · a source overrides it in its own card
Every 6 h ▾
2
Reconciliation schedule
Default reconciliation window full source hygiene · a source overrides it in its card 2
Frequency
Weekly ▾
Day
Sunday ▾
Time
03:00 ▾
↳ every Sunday at 03:00 · organization time (Europe/Moscow)
  1. 1 The embedding-model chip is read-only; it is taken from the AI models registry, and clicking it goes to the model selection there.
  2. 2 Global default collection cadences, which a source overrides in its own card: incremental — the freshness interval; reconciliation — how often a full sweep runs (for weekly, the day and time). The modes live in Harvester · sync-modes.
Sources

Connecting sources is optional at launch. Without them the knowledge base is empty (is_empty), and the platform runs as a pure AI assistant — chat and agents answer from the model's general knowledge. An assigned chat model is enough; sources add grounding on company data once the company is ready for it. The first accepted chunk switches grounding on by itself.

source list
achilles.local/admin/harvester /admin/harvester · Owner / Admin
Sources ⟳ Sync all 3 Add source 4
Source ↕ State ↕ Health ↕ Last sync ↓ 5 Entities ↕
Jira · Acme 6 Active syncing · 62%
running now 8.2k of 13.1k
48.1k ⋯ 7
Confluence · Acme Active idle
2 h ago in 4 min · +1.2k 3 in DLQ 8
61.3k ⋯
Slack · acme.slack.com Paused 9 idle
yesterday in 11 min · +3.4k
33.2k ⋯
GitLab · git.acme.dev Disconnected error 10
3 days ago credentials error
8.4k ⋯ 11
⋯ source · idle
Sync Pause Open card
⋯ source · syncing
Sync run in progress Pause run in progress
Cancel run Open card
⋯ source · disconnected
Reconnect Open card
Purge data
3 in DLQ · triage
3 in DLQ
No permission 2 Malformed object 1
↻ Retry failed
  1. 3 “Sync all” runs an incremental over every Active source at once; Paused and Disconnected are skipped, and with nothing to run the button is disabled. The mechanics live in Harvester.
  2. 4 “Add source” opens the connection wizard.
  3. 5 Sorting — click a header; a second click flips the direction, and the active column is highlighted.
  4. 6 Clicking a row opens the source card.
  5. 7 The row's ⋯ menu: its contents depend on the state. During a run only “Cancel run” is available; the rest is disabled with the reason in the tooltip.
  6. 8 The run finished with errors: some items failed to load and went to the DLQ. The chip opens a breakdown by cause; “Retry failed” runs only those.
  7. 9 Two axes per source. State — the admin's decision: Active · Paused · Disconnected. Health — the platform's observation: idle · syncing · error. The full axes live in Harvester.
  8. 10 error refers to the last run and does not block the configuration — it is fixed with a retry; a failure raises a notification.
  9. 11 A Disconnected source's data is orphaned but stays in the Knowledge Store. “Purge data” irreversibly removes the source's entire contribution, behind type-to-confirm.