Phase 2C — Bulk Facade: CLI ai:ingest-tenant + Streaming Ingestion Path

Context

Sub of Phase 2 Epic #11626 (meta-Epic #11624). Graduated from Discussion #11623.

Bulk data-plane facade for initial tenant onboarding + hook bursts > MCP threshold. CLI path bypasses MCP volume gate.

The Problem

Initial tenant onboarding can be 5k-50k chunks (per ChromaDB defrag-perf anchor: ~10k chunks reasonable scale). MCP facade refuses this. Hook bursts after git push of a large refactor branch can also exceed the threshold. Bulk path is structurally necessary.

The Fix

CLI command: npm run ai:ingest-tenant <tenantId> [--from-file <path>] [--from-stdin]

Wraps KnowledgeBaseIngestionService.ingestSourceFiles (Phase 2A) with viaMcp: false:

1. Reads parsed-chunk-v1 records from file OR stdin (JSONL streaming)
2. Batches into reasonable chunks (config-tunable; default 500 records)
3. Streams progress to stdout
4. Returns structured summary on completion

Optional V1.5 — HTTP/streaming endpoint deferred decision pending V1 deployment shape. For V1: CLI-only acceptable; HTTP can land as separate ticket if measurement justifies.

Script location: buildScripts/ai/ingest-tenant.mjs OR ai/scripts/ingest-tenant.mjs — final placement requires structural-pre-flight skill consultation at implementation time (sibling-pattern lift: ai/scripts/orchestrator-daemon.mjs precedent).

Acceptance Criteria

• CLI command npm run ai:ingest-tenant defined in package.json
• Script implementation reads parsed-chunk-v1 records from --from-file OR --from-stdin
• Batches records (default 500/batch; configurable)
• Streams progress + final summary to stdout
• Bypasses MCP volume gate (viaMcp: false)
• structural-pre-flight skill consulted for script placement decision (buildScripts/ai/ vs ai/scripts/)
• Unit tests: stream parsing, batching, error handling, summary shape
• Integration test: ingest 1k+ chunks via CLI; verify Chroma state matches

Out of Scope

• HTTP/streaming endpoint → potentially V1.5 (separate ticket if needed)
• MCP facade → Phase 2B
• Service-layer ingestion logic → Phase 2A (consumed here)

Related

• Parent: #11626
• Blocked-by: Phase 2A (#TBD — service layer must exist)
• Daemon-script precedent: ai/scripts/orchestrator-daemon.mjs, swarm-heartbeat-daemon.mjs
• Discussion source: #11623 §4 Q3 + §7 Phase 2

Origin Session ID

7360e917-1733-4cdd-a6f3-5ac51c34b838

Handoff Retrieval Hints

• ai/scripts/orchestrator-daemon.mjs is sibling-pattern reference for CLI tooling under ai/scripts/
• buildScripts/ai/backup.mjs (#10129 precedent) is the canonical-orchestrator-script pattern for KB-side ops
• structural-pre-flight skill placement decision is the substrate-correct gate before file location commits