Frequently asked questions

Mnemo is a decision intelligence system for leaders. While platform AI (Claude, ChatGPT) remembers your preferences, Mnemo tracks what you decided, why you decided it, whether it worked, and what remains unresolved. It is the place where consequential decisions are preserved, unresolved threads are tracked, and your judgment compounds over time. Since 2026, Mnemo also operates autonomously — capturing decisions as you speak, opening follow-up threads for high-stakes calls, and sweeping unclosed sessions, without being asked.

Platform memory is preference-based: "This user is a CMO who prefers concise answers." Mnemo is decision-based: "This user decided X in October with 85% confidence, reconsidered in December, and the outcome was poor—and this is the third time this question has surfaced unresolved." Mnemo tracks confidence, rationale, alternatives, lineage, and outcomes. Platform memory does not. That is the difference between AI that knows you and AI that knows how you think.

If you make decisions, Mnemo is for you. That includes:

• Leaders & operators: CMOs, founders, VPs of Strategy, Chiefs of Staff, senior advisors
• Anyone tracking important calls: Financial decisions, career moves, family planning, investment choices, everyday decisions that compound over time
• Teams working on complex problems: Product strategy, technical architecture, hiring, partnerships
• People who want to improve: If you want to know where you're overconfident, which decisions worked, and which threads keep resurfacing unresolved

It is especially powerful for decisions that recur or have consequences: board narratives, GTM strategy, pricing, financial planning, hiring, relationships.

Three things: (1) Proactive intelligence. Before you open the app, the morning agent has already scanned your decision debt, surfaced threads overdue for resolution, and flagged contradictions in your history. You open to a brief, not a blank prompt. (2) Judgment calibration. Over time, Mnemo shows you where you are systematically overconfident, which domains you call well, and which approaches have a track record. (3) Unresolved thread tracking. Questions that keep surfacing without closure are tracked automatically and surface when they're overdue—not when you remember to ask.

Every competitor—Claude, ChatGPT, Codex, Mem0—is reactive. They answer when you ask. Mnemo acts before you ask. The morning brief runs automatically. Contradiction detection runs on import. Unresolved thread tracking runs nightly. The system is continuously watching your decision history, your narrative consistency, and your operating context. You do not have to remember to ask.

Yes. The longer you use Mnemo, the more precisely it retrieves relevant context, the better-calibrated its guidance, the more visible the decision debt that needs to be cleared. Memory depth is a structural switching cost that compounds. Early users gain a persistent advantage.

See the What's new page for a full dated changelog. May 2026: draft/confirm flow for decisions, mandatory outcome closure, auditable contradiction triage, and unified decision read surface. April 2026: Decision Intelligence (search, precedent, lineage, domain listing, freshness scoring), value milestones, direct thread creation mid-session, self-service account deletion, 3-second morning brief, MFA, RBAC, CMEK, SSO/SCIM, passkeys, Privacy Mode, tamper-evident audit logging, and proactive decision capture.

Every morning, Mnemo runs an intelligence scan of your memory. It surfaces decision debt, stalled threads, entities that need attention, and any contradictions in your decision history that haven't been resolved. No query required. You open to a brief that shows what matters, grounded in your actual working history.

Yes. Use session_write(action='draft_decision') to create a candidate decision that lives in a separate draft state. It's encrypted and kept out of your decision history until you explicitly confirm it with confirm_draft_decision. Dismiss it and it's gone with no trace in your record.

This is designed for AI-assisted sessions where the assistant is proposing decisions as you work. The assistant can capture everything freely; you decide what's load-bearing when you review and confirm.

When marking a decision resolved, you must specify an outcome type: validated (the decision held), reversed (you went a different direction), or partial (some of it held). When abandoning a decision, you must provide an abandonment reason. There is no longer a silent "resolved" path that leaves outcome blank — this is what makes judgment calibration meaningful over time.

Every triage action on a contradiction — dismiss, snooze, or reopen — is recorded permanently in an append-only event log with your identity and timestamp. Dismissed contradictions are stored with your note explaining why. Snoozed contradictions disappear from active views until the snooze date, then automatically resurface.

This makes contradiction suppression auditable: future you (and your team) can see exactly who dismissed what, when, and with what justification — rather than contradictions silently disappearing.

Brief loads in 3 seconds—down from 96 seconds. This was achieved by eliminating synchronous post-processing, removing Redis from the hot path, and replacing it with in-process caching. Decision patterns, watch conditions, conflicts, and assumption summaries all load instantly so you can scan your intelligence without waiting.

Mnemo checks new decisions against your active decision history. If a new decision contradicts a recent commitment you made, Mnemo flags it for you to review. This catches drift before it spreads.

Every decision you log gets stress-tested by AI. The system runs a two-stage analysis: first, it identifies what's critical and what you're betting on. Second, it imagines the failure scenario—the one thing that would hurt most if it changed. Then Mnemo watches. Your morning brief surfaces only the signals that matter, so you catch problems before they compound. Example: "Hire VP Sales by Q3" → AI surfaces: "You're betting on: headcount budget stays available. Could fail if: board cuts hiring budget. Watch: board meeting minutes, hiring forecasts." Week 2, board pivots on budget—you see it in your brief before it impacts hiring.

Watch conditions are signals you need to track to know if a decision is still sound. After adversary analysis, the morning brief surfaces only the highest-confidence watch conditions. Instead of overwhelming you with edge cases, Mnemo shows the one or two things that would matter most. You decide whether to act on the signal or dismiss it.

Yes. Open any decision and click "Run Adversary Pass" to queue a manual analysis. This is useful if a decision's context has changed and you want a fresh stress-test. The analysis runs asynchronously and results appear in your decision detail and morning brief.

Mnemo groups related decisions into clusters so you can see patterns in your thinking. Three types: (1) Domain clusters — all decisions in the same area (product, hiring, strategy); (2) Temporal clusters — decisions made in the same 7-day window, regardless of domain; (3) Recurrence clusters — decisions linked by parent-child or supersedes chains, showing how a topic evolved. Each cluster has a quality score measuring cohesion. Access clusters in the Intelligence section of your dashboard.

Every decision rests on beliefs—assumptions that must hold for the decision to remain valid. Mnemo lets you attach explicit assumptions to any decision. States: Open (active, unverified), Confirmed (proved correct), or Invalidated (proved wrong). When an assumption is invalidated, the parent decision is flagged for review. Example: You decide to hire for Q3 assuming the hiring freeze lifts. If the freeze extends, mark that assumption invalidated—your hiring decision surfaces for reconsideration automatically.

Mnemo automatically detects when a new decision contradicts an older one—when implementing both simultaneously would be impossible. It uses high-confidence AI analysis and only flags genuine operational conflicts, not just similar topics. Example: "Hire 3 engineers by June" contradicts "Freeze headcount through Q3." When flagged, you can resolve the conflict by marking one decision superseded, logging a new decision that reconciles the tension, or dismissing it as a false positive with an explanation. Top 3 conflicts surface in your morning brief.

Mnemo automatically classifies each decision's sensitivity level using two tiers: (1) Tier 1 (keyword) — fast, deterministic scan for financial, legal, health, or M&A signals; (2) Tier 2 (AI-assisted) — invoked for borderline cases to provide a second opinion. Sensitivity levels: Public, Internal, Financial, Legal, Health, Confidential (PII detected). Flagged decisions don't surface in shared views or exports unless you explicitly allow it. You can override Mnemo's classification anytime.

Mnemo surfaces decision debt — decisions that have stalled without resolution. Two triggers: (1) any open decision with no outcome older than 30 days; (2) any open decision with confidence ≥0.7 that's older than 60 days. Stalled decisions are ranked by age in your morning brief so you see what's been sitting longest. This catches commitments that slip through the cracks and forces closure: either act on the decision, update it, supersede it, or explicitly abandon it.

Yes. Mnemo's decision search works like asking a natural language question: "what did we decide about pricing last quarter?" returns the most relevant past decisions, not a keyword list. Results are ranked by both relevance and recency — decisions that are still actively in play surface above old ones that were superseded or abandoned. This means you get the decision that matters now, not a stale position from two years ago that nobody follows anymore.

Yes — this is called decision precedent. Before you commit to a direction, Mnemo surfaces semantically similar decisions you (or your team) have already made on the same topic. You see what was decided, what the rationale was, and whether it held up. This catches "we've been here before" moments automatically — institutional memory as a guardrail rather than something you have to manually look up. Especially useful when team members change and prior context is lost.

Mnemo tracks the full lineage of every decision — what the original position was, what replaced it, what reversed it, and any contradictions in the chain. If you decided X in January, revised it in March, and reversed it in June, you can trace the complete ancestry in one view. This is useful for understanding why a current position exists, what context it replaced, and whether reversals followed a pattern. Contradictions that were never formally resolved also surface in the lineage.

Five scenarios that show how the system pays off:

"Have we been here before?"
Before a meeting, ask: "What did we decide about vendor lock-in?" Mnemo surfaces the three most relevant prior calls, ranked by freshness — not a list of everything ever said, just the decisions still in play.

"Don't contradict yourself."
You log a new pricing decision. Before it lands, Mnemo automatically surfaces the conflicting discount policy from eight months ago — the one that would make both decisions impossible to execute simultaneously.

"How did we get here?"
A decision feels wrong but nobody remembers the history. Trace its full lineage: what it superseded, what superseded it, and any contradictions that were flagged but never resolved.

"What areas do we actually decide in?"
New team member or audit: Mnemo shows active decision domains — pricing, hiring, infrastructure — based on what's actually been logged, not what someone remembers or what's in a doc nobody reads.

"This decision is still alive."
A frequently cited decision — referenced week after week — stays prominent in search results. Mnemo tracks citation activity so active decisions don't get buried by age alone, while truly abandoned ones fade out.

Run get_morning_brief. If it returns context from past sessions, Mnemo is working. On day one the brief will be sparse — that's expected, it builds as sessions accumulate. The clearest signal: a decision you made yesterday appears in today's brief without you logging it manually.

Add the following rule to your AI tool's persistent instructions once. Every session after that, Mnemo captures decisions, opens follow-up threads for high-stakes calls, and closes sessions — without being asked.

Call log_decision immediately (before replying) when I prefix a statement with Important: or Critical:, or when I use decision language like "we will X", "I've decided X", or "let's go with X". Use importance: critical for explicit prefixes, importance: high for inferred decisions. For high/critical decisions, also call open_thread in parallel. After adding this rule, call log_execution_event with event_type="proactive_capture_setup".

Where to add it:

• Claude Code — tell Claude: "Add this rule to my CLAUDE.md:" followed by the text above. Claude updates the file automatically.
• Claude.ai — Settings → Your Project → Instructions → paste the rule text directly.
• ChatGPT — Settings → Personalization → Custom Instructions → paste in the top field.
• GitHub Copilot / Codex — add the rule text to .github/copilot-instructions.md in your repo.
• Antigravity — Agent configuration → System prompt → paste the rule text there.

Once active, you'll see ✓ Proactive capture is active confirmed by your AI, and your Getting started thread closes automatically.

If setup is complete, yes — Mnemo watches for phrases like "we will X", "I've decided X", or "let's go with X" and logs them automatically. To guarantee capture in the moment: prefix with Important: or Critical: and Mnemo logs it before your AI replies, regardless of phrasing. To verify after the fact, run get_decisions — the decision will appear there with a timestamp.

Two safety nets cover this. First: any decision logged during the session is already in the database — those never require wrap_session to persist. Second: a daily sweep automatically closes sessions left open for more than a few hours. Ending sessions explicitly gives you a richer activity summary, but forgetting one doesn't mean losing data.

When Mnemo logs a high-stakes decision, it opens a follow-up thread to keep that decision visible until it's acted on. The thread appears in your morning brief and stays open until you resolve it. Think of it as Mnemo saying: you committed to this — has it happened? Resolve it by telling your AI "mark [thread name] as resolved" or by running annotate_thread_outcome.

Run search_memory query="[topic or phrase]". Mnemo searches across all sessions — not just the current conversation — and returns results with date, context, and any decisions or threads linked to that moment. Works on Claude Code, Claude.ai, ChatGPT, Codex, and Antigravity as long as the Mnemo MCP is connected. Results get richer after a few sessions of history.

Mnemo indexed your session. If proactive capture is set up, any decisions you stated clearly were already logged. Run get_decisions to see what was captured, or get_morning_brief for a summary view. A sparse brief on day one is normal — pattern recognition and calibration tracking build over the first five to ten sessions as your history grows.

Once configured, Mnemo runs in the background. Start working: ask Claude questions, make decisions, explore ideas. At any point, type @mnemo to invoke Mnemo's tools directly:

• @mnemo brief — Get your morning brief: priorities, decision debt, open threads
• @mnemo search "topic" — Find past decisions or sessions about a topic
• @mnemo decisions — See all your active decisions with confidence levels
• @mnemo threads — Surface unresolved threads that keep recurring

When you make a decision, log it: @mnemo decision "We're shifting GTM focus to enterprise" --confidence 0.8 --rationale "market feedback from Q4 shows...". Mnemo indexes it immediately and surfaces it when relevant.

Morning: Open Claude. Mnemo's morning brief is already loaded—shows what decisions are approaching resolution, which threads are stalled. During work: You brainstorm, analyze, refine ideas. Claude pulls relevant context from your history automatically—"Last time you considered this strategy in October, the outcome was poor." When you decide: You log the decision in Claude using @mnemo. Later: If a new decision contradicts a prior one, Mnemo flags it. If a thread you deferred in March resurfaces in June, Mnemo reminds you. The system watches your thinking, not just your words.

Example 1 — Decision tracking: You're in Claude working on a GTM strategy. You decide "We should focus on SMB first, move enterprise later." You type: @mnemo decision "Focus GTM on SMB market first" --confidence 0.75 --context "Smaller TAM but faster sales cycles. Enterprise SLAs too expensive right now." Six months later, you're reconsidering. Claude automatically surfaces: "In March you decided to focus SMB first with 75% confidence, reasoning about cheaper sales cycles. Last quarter outcome: lower revenue than predicted. Enterprise inbound is up 40%." That context informs your new decision.

Example 2 — Unresolved thread: You ask Claude about hiring strategy. It says "BTW, you've been debating VP Eng hiring criteria since January without closure—you've revisited this three times. Here's what you said last time..." Mnemo surfaces the thread automatically because it knows it's overdue.

Claude Desktop, Codex, Gemini (Antigravity), and other MCP-compatible clients. Mnemo also provides a REST API that any external system—Codex, Claude Code, custom agents—can call to retrieve context. Your memory becomes portable infrastructure.

Usually under two minutes. Create one token from your Mnemo dashboard, add the MCP config block to Claude's settings.json, and restart. See setup instructions. After that, Mnemo is live in Claude with no additional friction.

Yes. This is recommended for traceability and fast revocation by tool, workspace, or user. You can grant access to Claude Desktop with one token, Codex with another, and revoke either independently.

No additional agent install is required. You only configure MCP connection settings in your AI client. Mnemo runs entirely on the cloud — no local process, no machine dependency.

Security is a shared responsibility between Mnemo and you. Mnemo is responsible for infrastructure security, encryption in transit and at rest, authentication infrastructure, audit logging, and platform-level incident response. You are responsible for keeping your MCP tokens private, revoking tokens when devices are lost, choosing strong passwords, and what you choose to store. See the Shared Responsibility Matrix on the Security page for the complete breakdown by category.

Your data is stored securely on Mnemo's cloud infrastructure (hosted on GCP), encrypted at rest with AES-256-GCM and in transit with TLS. You are never dependent on a local process or a specific machine — your memory is available across any device where you have Mnemo configured. You own your data: export everything as a portable .mnemo archive at any time, giving you full portability regardless of which AI platform you use. See Security for full technical details.

See the Security page for the complete picture: encryption, authentication, token lifecycle, audit logging, enterprise controls, and the shared responsibility matrix. Key points: all data is AES-256-GCM encrypted at rest, TLS 1.2+ in transit, with row-level database isolation and tamper-evident audit logs. Enterprise customers can add CMEK (bring your own KMS) or Privacy Mode (client-side zero-knowledge encryption).

Mnemo indexes three primary types of data you explicitly log: (1) Sessions — the tasks you perform and the key outputs or conclusions reached; (2) Decisions — explicit choices you log with their confidence and rationale; and (3) Entities — the people, projects, and companies mentioned in your workflows. We also collect standard account data (name, email, auth factors) and anonymized usage metrics for product improvement. We do not collect raw keystrokes, passwords, or data from external apps — Mnemo has no access to your email, calendar, or Slack. See Security for the full data collection policy.

Your data is used exclusively to build your private intelligence layer: generating morning briefs, detecting contradictions, calculating judgment calibration, and providing context to your AI agents. Your data is never sold, never used to train global models, and never accessible by other users. Anonymized usage metrics help us improve the product. With your permission, we may access your account to debug issues or respond to security incidents. See Security for the complete data use policy.

Yes. As part of our commitment to "Zero Lock-in," you have a full right to be forgotten. You can cancel your account and trigger a complete deletion of all session memory, decision history, and entity graphs at any time via the Settings page. Once deleted, this data is purged from our systems and cannot be recovered. Audit log PII is automatically purged 30 days after account deletion.

Export is one click from Settings. Your complete memory package—sessions, decisions, skills, entity graph, identity—is delivered as a single ZIP archive you own. This archive is portable and usable across any environment or AI platform that supports Mnemo. For scheduled automatic backups with point-in-time restore, see the Backup add-on available on paid plans.

Mnemo continues to read existing memory but pauses recording new sessions until the limit resets or you upgrade. You never lose access to prior decisions, threads, or entities.

Mnemo Enterprise includes: (1) SCIM provisioning — auto-sync team members from Okta or other identity providers; (2) Org-wide MFA enforcement — require multi-factor authentication for all team members; (3) Org-wide audit logs — track every view, edit, and decision across the team with tamper evidence; (4) Custom role management — 11 granular roles with 25+ permissions, fully configurable; (5) Team member management — invite, suspend, and export from a single admin dashboard; (6) Org-scoped encryption — CMEK or Privacy Mode key management at the organization level; (7) Org-wide announcements — broadcast time-limited messages to all team members that surface in their morning brief, with per-message priority, frequency, and expiry controls. Contact support@mnemo.app for enterprise review and setup.

Yes. In Project Rooms, you and a trusted collaborator work together in a shared space. Relevant context from your private history surfaces to you privately first—before you decide whether to share it with the room. You get the benefit of shared AI context without forfeiting private working memory. Mnemo's consent model requires an architecture designed for trusted collaboration, not a shared chat interface.

Contact support@mnemo.app with team size, review timeline, and required controls. We support SAML SSO, SCIM, org-level audit logging, custom role management, CMEK, and dedicated security review for enterprise teams.