Brain: Persistent Memory
Your AI sessions contain decisions, commitments, patterns, and context that matter. 3ngram captures these as structured memory types with lifecycle semantics, not flat text.Commitments
Things you said you’d do. Due dates, status tracking (open, waiting, scheduled, resolved), overdue detection.
Decisions
Choices you’ve made and why. Searchable across sessions so you don’t relitigate.
Blockers
What’s in your way. Tracked until resolved, surfaced in briefings.
Context
Background facts, patterns, preferences. Anything that doesn’t fit a stricter type.
Scope and project labels
Every memory can be taggedwork or personal, with an optional project name. Filter commitments by project, recall only work decisions, or scope a briefing to one project.
Status tracking
Commitments move throughopen → waiting → scheduled → resolved. 3ngram auto-detects overdue and stale items so nothing slips quietly.
Auto-consolidation
When you save a memory that closely matches an existing one, 3ngram merges them instead of creating a duplicate. No manual dedup needed.Entity extraction
3ngram extracts named entities (people, organizations, projects, tools, concepts) from every memory at save time. Entity chips appear in search results and the Cmd+K command palette — no setup needed.MCP tools
Connect 3ngram to Claude, ChatGPT, Cursor, or any MCP-compatible AI tool. Your context follows you between clients.| Category | Tools |
|---|---|
| Memory | remember, search_memories, reclassify_memory, batch_reclassify, archive_memories |
| Accountability | resolve, unresolve |
| Search and index | search_content, reindex |
| Organization | configure_scope, reapply_scope_mappings |
| Content sources | sync_source |
| Google Docs | docs_create, docs_export |
| Actions | propose_action, decide_action |
| Preferences | set_preference |
| Custom prompts | create_prompt, update_prompt, delete_prompt, get_prompt |
| Schedules | create_schedule, update_schedule, delete_schedule, get_schedule_history |
| Triggers | create_trigger, update_trigger, delete_trigger, get_trigger_history |
commitments, recall, handoff) still work but redirect to the corresponding resource or active tool.
All tools work in plain language. No function syntax needed.
Full tool reference
Parameters, read/write classification, and usage examples for every tool.
Prompt templates
Ten built-in prompt templates orchestrate multiple tools into repeatable workflows.| Prompt | What it does |
|---|---|
| Briefing | Start-of-session overview: commitments, blockers, overdue, stale, recent decisions |
| Debrief | End-of-session extraction: scans conversation, persists decisions and commitments, resolves completed items |
| Project context | Deep dive into one project: memories, commitments, blockers, decisions |
| Meeting prep | Pre-meeting context: upcoming events, related memories, open commitments |
| Post meeting | Post-meeting capture: decisions, action items, blockers from the meeting |
| Compile context | Aggregate all relevant context for a topic or project into a single briefing |
| Handoff | Export session context for transfer to another AI tool or collaborator |
| Triage stale | Review and act on commitments that have gone stale |
| Suggest reclassifications | Identify memories that may be miscategorized and propose corrections |
| Onboarding | Guided first-session walkthrough for new users |
Prompt templates guide
Usage patterns, client compatibility, and invocation examples.
Knowledge: Search Your Work
Memory is what you have decided and committed to. Knowledge is the full context of your work. 3ngram connects to the tools where your work already lives and makes everything searchable from inside your AI chat.- GitHub repos for architecture, PRs, and decisions in code
- Google Drive and Google Docs for proposals, contracts, and specs
- Jira, Linear, Basecamp, Confluence for tasks, issues, and project context
- Google Calendar for meetings, deadlines, and scheduling context
- Local markdown for notes, SOPs, and working documents
- Semantic search from MCP tools inside your AI chat, with synthesized answers and source citations
- Cmd+K search across everything: memories, documents, entities, commitments
- Knowledge graph mapping relationships between people, projects, tools, and concepts
- Handoff export when you need to switch AI providers or sessions
Chief of Staff: Proactive Surfacing
A brain without follow-through is just a filing cabinet. 3ngram does not wait for you to ask.- Overdue detection on commitments past their deadline
- Stale alerts on commitments with no activity past a configurable staleness threshold
- Daily digest emails with overdue items, stale commitments, and recent captures
- Nudges in the dashboard: “You committed to X, have you started?”
- Snooze an alert for 1, 3, or 7 days
Agent automation
Schedule prompts to run on a recurring basis, or trigger them when events happen.Schedules
Run custom prompts on a cron schedule. Automate daily briefings, weekly reviews, or periodic check-ins.
Triggers
React to events automatically. Fire prompts when commitments become overdue, blockers appear, or integrations sync.
Actions
Let your AI propose write-back actions on GitHub and Linear. Review and approve before execution.
Recurring commitments
Commitments that auto-repeat. Resolve one, and the next occurrence appears with an updated due date.
Dashboard
A web UI for reviewing and managing follow-through without opening an AI client.| View | What you see |
|---|---|
| Summary | Counts for memories, commitments, decisions, documents, clusters |
| Commitments | Open items with overdue and stale indicators, filterable by scope and project |
| Nudges | Proactive prompts with dismiss and snooze controls |
| Alerts | Overdue and stale commitments grouped by urgency |
| Insights | Completion rate trends, active day streaks, resolution velocity |
| Decisions | Recent decisions log |
| Clusters | Automatically grouped memories for spotting patterns |
| Daily note | Today’s note with previous-day navigation |
| Analytics | Memory type distribution, commitment health score, tool usage stats |
Document editor
- Rich text editor (Tiptap) with autosave
- Folder-based organization
- Full-text search with relevance scoring and snippet highlighting
- Unified search across documents and memories (Cmd+K)
Import and export
Import your existing context from:- ChatGPT conversation exports (JSON)
- Claude conversation exports (JSON)
- Craft markdown files (
.mdor.zip)
Automatic capture (Claude Code)
Claude Code hooks auto-capture context without manual debriefs:| Hook | What it captures |
|---|---|
| Session start | Loads your briefing into session context |
| Post tool use | Captures git commits as memories |
| Session end | Persists files modified and commits as a summary |
source=hook so they don’t mix with deliberate saves.
Hook setup guide
Installation, environment variables, and capture filter configuration.
Cross-tool continuity
Memories saved in one AI tool show up in every other connected client. Switch between Claude Desktop, Claude Code, Cursor, and ChatGPT without losing context.Connect your AI tools
Step-by-step setup for each client.
Hosted with a free Starter tier
The primary product experience is hosted atapp.3ngram.ai. Sign up for a free Starter account to get started, no credit card required. Upgrade to Individual for unlimited memories and content sources.
Get started
Sign up for a free Starter account.
Plans and billing
Plan comparison, memory limits, and add-on pricing.
Privacy and data control
- Row-level security enforces strict user isolation at the database layer
- Audit trail of admin actions
<private>tags let you exclude sensitive content from storage- Export everything at any time, regardless of account status
- LLM consent is opt-in: memory consolidation via external LLM defaults to off