Loading live MongoDB capabilities…

Vendor integration

MongoDB Atlas

Persistent agent memory · multi-store audit trail · vector + lexical retrieval

Status

● Live

production · real data flowing

— trade decisions · — collections · — indexes — persistent agent memory

Capability matrix

What we use, what's gated, what's planned

The persistent memory of the agent stack. the full trade decisions across many collections — every evaluation, outcome, and reflection the platform has made — let agents reason against the operator's full history, not just the current market frame.

Coming nextAtlas Vector Search powers semantic retrieval over the decision corpus; multi-tenant collection partitioning unlocks the Phase 2.5 SaaS surface.

CapabilityStatusPhaseWhy / How / Note

Operational document store

LIVEPhase 1

6 collections persist every decision, outcome, snapshot, and agent reflection.

95 tests

Aggregation pipelines

LIVEPhase 1

$group + $match + $sort power the /insights flywheel cuts.

Replace-one upserts

LIVEPhase 1

Idempotent writes via deterministic IDs — daemon can replay without duplication.

Distinct facets

LIVEPhase 1

/decisions/facets endpoint powers the faceted UI.

Atlas Search Lucene FTS

LIVEPhase 1.5.S (S2)

BM25 ranking on reasoning corpus — replaces regex with relevance scoring. M0-tier free.

16 tests

Schema validators on annotations

LIVEPhase 1.5.S (S1)

Atlas-side enforcement of label/score range; impossible to write malformed data.

Atlas Vector Search

CODE READYPhase 1.5

768-dim semantic search over reasoning corpus via gemini-embedding-004.

Implementation merged; M0 → M10 cluster upgrade required for vector indexes (~$60/mo).

Time Series Collections

PLANNEDPhase 2.5

Auto-bucketed storage for market_snapshots — 10× compression past 100K rows.

market_snapshots is the candidate; migration script TBD.

Change Streams

APP-014

PLANNEDPhase 2.0

Reactive embedder fan-out — new decision → embed → trace, no polling.

Replaces the current Cloud Scheduler poll loop.

Atlas App Services Triggers

UNDERUSEDPhase 2.5

Server-side serverless on change streams — cheaper than Cloud Scheduler.

Online Archive

UNDERUSEDPhase 3

Tier old decisions to cold storage — needed past 100K decisions for cost flatness.

VPC Peering with GCP

PLANNEDPhase 2.5

Eliminates the 0.0.0.0/0 allowlist; static private IP for FCA audit.

Eliminates the broad IP allowlist; required for FCA-grade security posture.

BYOK encryption (GCP KMS)

APP-018

PLANNEDPhase 3

Customer-managed keys — FCA-relevant for live-capital phase.

Live data

Real-time from the MongoDB Atlas backend

Every tile below is a live read from the vendor backend via the FastAPI BFF. If a tile shows "—" the backend is unreachable or the metric is not yet wired (no hardcoded numbers — see anti-pattern #2).

Awaiting backend

Awaiting backend

Awaiting backend

Awaiting backend

Roadmap commitments

Roadmap dependencies

Capabilities enabled by this integration — what is built, what is gated, and why.

APP-012Phase 2.5.1

Multi-tenant data partitioning (shared collections + tenant_id field)

Atlas schema-validator + middleware + integration test triple-enforces tenant isolation

APP-014Phase 2.0

Per-tenant LLM budget circuit breaker

tenant_metadata collection holds per-plan token caps; reactive enforcement

APP-015Phase 2.5

GDPR DSR cascade (Article 17 right-to-erasure)

Mongo is one of 5 backends in the cascade; per-tenant deletion via tenant_id filter

APP-017Phase 2.5.5

Vector-search tenant filter mandatory

Pydantic wrapper + integration test prevents cross-tenant leak in $vectorSearch

Demo flow

End-to-end showcase journey

Five steps a judge or investor can replay live. Each step links to the page that demonstrates it.

  1. 1

    Open /portfolio. The 'MongoDB Memory' card shows 4,251 docs across 6 collections — live. → Open /portfolio

  2. 2

    Click 'Query Atlas' on the MongoDB Memory card. Type 'FOMC LOSS' → semantic search via Atlas Search Lucene returns 12 ranked hits in under 200ms. → Open /insights

  3. 3

    Open /decisions, type 'breakout failed in tech'. Source badge shows 'atlas-search' — BM25 ranking, not regex. → Open /decisions

  4. 4

    Click 👎 on a result. Annotation lands in the new decision_annotations collection (Atlas schema-validator enforces label/score range). → Open /decisions

  5. 5

    Open /insights. Self-Improvement Loop widget shows aggregate annotation counts + top-5 lowest-approval strategies, all backed by Mongo aggregations. → Open /insights

What's next

Top-3 vendor-enabled capabilities coming soon

Sourced from the vendor's playbook. Each entry is mapped to its delivery phase and the value it unlocks.

Time Series Collections

Phase 2.5

Storage cost flatness past 100K market_snapshots

Change Streams

Phase 2.0

Reactive embedder fan-out; retire polling

Online Archive

Phase 3

FCA evidence retention beyond 1 year

AI Strategy Advisor

agents · tools · Gemini Flash

Welcome to SentinelHub. I'm the AI Strategy Advisor — I coordinate 9 specialist agents to help you analyse markets, build strategies, and manage risk.

Ask me about market conditions, your portfolio, trading strategies, or past performance.

Powered by Gemini Flash · Connects to Alpaca, MongoDB, Phoenix