Architecture

Arc AI active

Two layers, one platform: pCloudy's real device cloud as the foundation, Arc's product-aware AI as the intelligence. One ingest path, one store, two consumers.

Sources

pCloudy

pcloudy

Completion event stream · Real-time

TEST_RUN, FAILURE

JIRA

jira

REST API delta poll · Every 15 min

BUG, STORY, TASK

GitHub

github

REST API delta poll · Every 30 min

PR, COMMIT

Documents

document

User upload · On demand

DOC, DOC_CHUNK

Store

Ingest Workers

stable_uuid dedup · embed · relationship link
T1 exact + T2 semantic cross-source links

pgvector · PostgreSQL

objects · relationships · embeddings · context_log

90-day rolling window · project_id isolated

Consumers

Ask Arc

NeoBank QA team · plain-English queries

Context API

Arc · agents that act, not just report

Feedback loop — every run's results flow back and enrich the context used for the next release decision.

Single storage layer for all sources

All four sources write into the same objects and relationships tables. The source field identifies origin; embedding (pgvector) enables semantic search across all sources in a single query. One operational system handles everything in v1.