ingest

Filesystem mode (server-local paths)

1. If the user passes --status, call mori-ingest_status and present the table.
2. If the user passes --preview or --dry-run with no source provided, remind them to provide --source <path>.
3. For --preview: call mori-ingest_preview with the --source path(s) and any --type or --since arguments. Present the chunk breakdown and cost estimate. Remind the user this is zero-cost — no LLM was called.
4. For --dry-run: call mori-ingest with dry_run=true and all provided arguments. Report what would be written and the actual cost incurred (the LLM was called, just nothing committed).
5. For a real ingestion run: call mori-ingest with all provided arguments. Report: sources processed, chunks sent, memories written, cost estimate, errors.
6. If errors occurred for specific files, report them individually.
7. After successful ingestion, suggest running /brief to reload shared memories so the new entries are visible.
8. For large or expensive-looking sources, suggest /ingest --preview first to check before committing.

Content mode (remote clients — HTTP upload)

When the user passes --content <path>, upload the file directly to the mori-ingestion pod. Do NOT use the mori-ingest_content MCP tool — it has been removed. Use HTTP multipart upload only.

Requires:

• MORI_INGEST_URL env var (set by installer — NOT derived from MORI_URL)
• MORI_API_KEY env var (same key as MCP auth)

Linux/macOS

JOB=$(curl -sf -X POST "${MORI_INGEST_URL}/api/ingest/upload" \
  -H "X-Api-Key: ${MORI_API_KEY}" \
  -F "files=@${path}" \
  -F "focus=${focus}" \
  -F "tier=${tier}" \
  -F "tags=${tags}" \
  -F "dry_run=${dry_run}")
JOB_ID=$(echo "$JOB" | python3 -c "import sys,json; print(json.load(sys.stdin)['job_id'])")

Windows (PowerShell)

$r = Invoke-RestMethod -Uri "$env:MORI_INGEST_URL/api/ingest/upload" `
  -Method POST `
  -Headers @{"X-Api-Key" = $env:MORI_API_KEY} `
  -Form @{
    files   = Get-Item $path
    focus   = $focus
    tier    = $tier
    tags    = $tags
    dry_run = $dry_run
  }
$JOB_ID = $r.job_id

Polling (both platforms)

Poll every 5 seconds until status is complete or failed:

# Linux/macOS
while true; do
  STATUS=$(curl -sf "${MORI_INGEST_URL}/api/ingest/job/${JOB_ID}" \
    -H "X-Api-Key: ${MORI_API_KEY}")
  STATE=$(echo "$STATUS" | python3 -c "import sys,json; print(json.load(sys.stdin)['status'])")
  [ "$STATE" = "complete" ] || [ "$STATE" = "failed" ] && break
  sleep 5
done
echo "$STATUS"

# Windows
do {
  $status = Invoke-RestMethod -Uri "$env:MORI_INGEST_URL/api/ingest/job/$JOB_ID" `
    -Headers @{"X-Api-Key" = $env:MORI_API_KEY}
  if ($status.status -notin "queued","running") { break }
  Start-Sleep 5
} while ($true)
$status

Report: memories written, estimated cost, errors. After success, suggest /brief to reload shared memories.

Error handling

• MORI_INGEST_URL not set → tell user to set it (e.g. http://localhost:8969) and re-run
• HTTP 413 → file too large (50MB/file, 200MB total)
• HTTP 401 → API key mismatch — check MORI_API_KEY
• status failed → report errors field from job response

Argument mapping

User flag	Action
--source <path> (repeatable)	mori-ingest MCP tool → source list
--content <path>	HTTP multipart upload to MORI_INGEST_URL
--type <type>	type (auto, transcripts, git, docs, image)
--focus <area>	focus (all, decisions, architecture, conventions, gotchas)
--tier <tier>	tier (working, canonical, ephemeral)
--tags <tags>	tags (comma-separated string)
--since <duration>	since (e.g. "30d", "90d") — filesystem mode only
--dry-run	dry_run=true
--force	force=true
--max-cost <amount>	max_cost (float, USD) — filesystem mode only
--preview	call mori-ingest_preview instead of mori-ingest
--status	call mori-ingest_status