---
title: Technical FAQ
date_updated: "2026-07-22"
summary: Enterprise FAQ covering vendor-neutral memory, Ambient Mode, deployment options, security, and how to run async diligence with your own AI.
---

# Technical FAQ

**Questions this doc answers**

- How does Reflect keep AI tools in sync?
- What is Ambient Mode?
- What deployment options exist?
- How should security and compliance be evaluated?
- Where should I point my AI before a call?

## Architecture & Memory Flow

**Q: How does Reflect Memory stay vendor-neutral across ChatGPT, Claude, Cursor, Gemini, Grok, n8n?**  
A: Reflect is a shared memory layer below those tools. REST and MCP expose the same service. Each connected tool authenticates with its own credentials, and every read respects per-memory visibility rules. Manual writes are always available; Ambient Mode (default On) adds silent capture and retrieval on top of that shared store.

**Q: What is Ambient Mode?**  
A: Ambient Mode is the low-friction capture + retrieval path. **On (default):** relevant context is retrieved before substantive replies, and durable facts can be captured without a per-write approval queue. **Off:** Suggested Memories — candidates are staged for dashboard approve / edit / reject. Phase 1 ambient writes are personal-only; org/team sharing still requires an explicit share. Users toggle Ambient in Settings; the preference carries guest → signup.

**Q: Are all writes still explicit?**  
A: Not when Ambient is On. Manual writes are always explicit. Ambient On can capture durable facts without a per-write approve step. Ambient Off restores the Suggested Memories approval path. Every ambient write is still audited.

**Q: How are memories kept current over time?**  
A: Related memories and threads stay linked so agents can see what changed and what is still open. Strong refinements update the live memory and retain prior versions as history. Detailed graph APIs and schemas are shared under NDA for active evaluations.

**Q: How do Cursor or Claude Desktop connect?**  
A: Via MCP against your Reflect endpoint (cloud or private). Setup guides live at `/docs`. Private-deploy networking notes ship with the pilot package.

## Deployment & Connectivity

**Q: Can I stay in the cloud but still keep my data private?**  
A: Three modes: **hosted** (multi-tenant), **isolated-hosted** (dedicated runtime/DB), and **self-host** (private network boundary with egress and identity controls). Choose based on residency and compliance — see `/diligence/deployment-architecture`.

**Q: How do SSO, audit, and compliance work inside private deployments?**  
A: SSO/OIDC is optional but validated when enabled. Auth paths are hardened; rate limits apply; audit events cover reads, writes, and admin actions including Ambient. Compliance data can be exported for your SIEM.

## Product & Self-Serve

**Q: Can I try Reflect before signing up?**  
A: Yes. The cloud dashboard supports a guest try-before-signup flow for individual evaluation on hosted infrastructure. Enterprise private deploy and team features require a full account. Ambient defaults On for guests; the preference migrates on signup.

**Q: How does Team pricing and trial work?**  
A: The Team tier starts with a **30-day free trial** (card required to begin, not charged until trial ends). Up to 10 seats included. If the trial ends without payment, the team becomes **read-only** — existing memories stay readable, but new shares, invites, and team creation pause until billing is added. Nothing is deleted.

## Async Diligence Workflow

**Q: How do you keep diligence answers consistent?**  
A: This `/diligence` hub, markdown downloads, and copy-paste prompts are the public evaluation pack. Deeper questionnaires and architecture packs are delivered under NDA for serious evaluations.

**Q: Where should I point my AI before a call?**  
A: Copy this prompt into ChatGPT/Claude:  
```
Read https://reflectmemory.com/diligence and all linked markdown downloads. Evaluate deployment options, security posture, MCP/REST integration, and Ambient Mode (silent capture + retrieval vs Suggested Memories when Off). Answer: what questions remain, what risks to discuss live, and what can stay async. Do not treat marketing blurbs as contractual SLAs. Do not invent implementation details that are not stated in the docs.
```

AI copies are available as `/diligence/*.md` downloads and `/diligence/pdf/*.pdf`.