Unlimited organized memory for your AI agent. Store, search, and organize projects, contacts, decisions, and knowledge across categories. Never lose context...
Memory
augmentiq
Try itAugmentiQ memory system — recall, reason, record, and consolidate memories through the AugmentiQ MCP server. Gives the agent persistent, evolving memory across sessions.
What it does
AugmentiQ memory system — recall, reason, record, and consolidate memories through the AugmentiQ MCP server. Gives the agent persistent, evolving memory across sessions.
The skill document
AugmentiQ 🧠 — Your AI's Memory. 100% Local. Yours.
Your AI forgets everything. Every conversation starts from zero. AugmentiQ fixes that — permanently.
What Is This?
AugmentiQ gives AI agents persistent, evolving memory that lives inside your Obsidian vault. No cloud. No subscription. No data leaving your machine. Your AI remembers your preferences, your projects, your decisions, and your working style — and gets smarter every time you use it.
┌──────────────────────────────────────────────────────────────┐
│ YOUR MACHINE │
│ │
│ ┌──────────────┐ ┌──────────────┐ ┌──────────────┐ │
│ │ Your Vault │◄──►│ AugmentiQ │◄──►│ MCP Server │ │
│ │ (notes + │ │ Plugin │ │ :3710 │ │
│ │ .augmentiq │ │ │ │ (local only) │ │
│ │ memory) │ │ Memory Loop │ │ │ │
│ └──────────────┘ └──────────────┘ └──────┬───────┘ │
│ │ │
└─────────────────────────────────────────────────┼──────────┘
│ MCP
(loopback only — no internet)
│
┌───────────────────────┼───────────────┐
│ │ │
▼ ▼ ▼
┌─────────────┐ ┌─────────────┐ ┌─────────────┐
│ OpenClaw │ │ Claude │ │ Cursor / │
│ + Skill │ │ Desktop │ │ Cline / │
│ (full loop)│ │ (MCP) │ │ Any MCP │
└─────────────┘ └─────────────┘ └─────────────┘
Everything stays on your machine. The MCP server binds to 127.0.0.1. No remote connections. No cloud calls. Your vault is the store. Your agent's LLM is the reasoning engine. No separate server, no external database, no subscription.
The Memory Loop: How It Works
AI agents don't naturally remember. AugmentiQ teaches them a human-like memory cycle:
┌─────────────────────────────────────────────────┐
│ │
│ 1. RECALL 2. REASON │
│ Before responding Apply recalled context │
│ Search vault + Deductive, inductive, │
│ memories abductive reasoning │
│ │
│ ────────────────────────────────────────── │
│ │
│ 3. RESPOND 4. RECORD │
│ Answer naturally Save what was learned │
│ Informed by what Type, scope, importance, │
│ was recalled tags — classified memory │
│ │
│ ────────────────────────────────────────── │
│ │
│ 5. CONSOLIDATE (periodic "dreaming") │
│ Review, deduplicate, summarize, │
│ promote patterns, archive stale entries │
│ │
└─────────────────────────────────────────────────┘
This is what makes AugmentiQ memory, not just retrieval. The agent doesn't just find text — it reasons about what it found, decides what's worth keeping, and organizes knowledge so it's useful later.
Why AugmentiQ, Not Something Else?
| RAG | Cloud Memory | Semantic Vault MCP | AugmentiQ | |
|---|---|---|---|---|
| Memory classification | ❌ | varies | ❌ | ✅ type, scope, importance |
| Audit trail (hash-chained) | ❌ | ❌ | ❌ | ✅ |
| Consolidation ("dreaming") | ❌ | ❌ | ❌ | ✅ |
| Your data stays local | ❌ | ❌ | ✅ | ✅ |
| Agent LLM is the reasoner | ❌ | varies | ✅ | ✅ |
| Works with ANY MCP client | ✅ | ❌ | ✅ | ✅ |
| Free & open source | varies | ❌ | ✅ | ✅ |
| Reasoning loop (deduce/induce/abduce) | ❌ | varies | ❌ | ✅ |
RAG retrieves text but doesn't reason about it. Cloud memory owns your data. Semantic Vault MCP gives raw access but no memory layer. AugmentiQ fills the gap: reasoning intelligence + vault-native + structured memory. All local. All yours.
Quick Start
1. Install the Obsidian Plugin
Search "AugmentiQ" in Obsidian Community Plugins, or use BRAT. Enable it in Settings → Community Plugins.
2. Connect Your Agent
OpenClaw:
openclaw skills install augmentiq
Claude Desktop:
{
"mcpServers": {
"augmentiq": {
"url": "http://localhost:3710/mcp",
"headers": { "Authorization": "Bearer ***" }
}
}
}
Cursor / Cline / any MCP client: Same URL, same auth header.
3. Ask Your Agent
"What do you know about me?"
Your agent will onboard, create its memory, and start remembering. Every conversation after that, it recalls what it learned and builds on it.
The Tools
5 tool groups, 20+ actions. All running locally.
| Tool | What It Does |
|---|---|
🧠 memory | Recall, remember, forget, health, profile, consolidate — the differentiator |
📁 vault | Read/write/search notes in your vault |
✏️ edit | Fine-grained editing with fuzzy matching |
🕸️ graph | Traverse links, backlinks, forwardlinks, path finding |
⚙️ system | Server info and diagnostics |
Security
- Bearer token auth — auto-generated, no key no access
- Loopback only —
127.0.0.1, no remote connections - Read-only mode — lock the vault for agents that shouldn't write
- Hash-chained audit trail — every memory operation is logged, tamper-evident
- You own the data — everything in your vault, no cloud, no subscription, ever
What This Skill Does (For Agents)
The sections below are instructions for AI agents using AugmentiQ. If you're a human, you can stop here — the Quick Start above is all you need.
When to Use
Use AugmentiQ in every main-session conversation with the user. The memory loop should be as natural as a human's memory: recall what you know, use it to inform your response, and remember what you learned.
Do not use AugmentiQ in:
- Group chats or shared contexts (memory is personal to the user)
- One-shot subagent tasks (no need to recall memories)
- Read-only operations that don't involve user interaction
Connection
The AugmentiQ MCP server runs locally alongside the Obsidian plugin.
- URL:
http://localhost:3710/mcp - Auth: Bearer token (read from
AUGMENTIQ_TOKENenv var or OpenClaw config) - Protocol: MCP over HTTP
Call tools via mcporter:
# Set up auth (once)
export AUGMENTIQ_TOKEN=""
# Call any AugmentiQ tool
mcporter call http://localhost:3710/mcp. --auth "Bearer $AUGMENTIQ_TOKEN" --args ''
# Example: recall memories
mcporter call http://localhost:3710/mcp.memory --auth "Bearer $AUGMENTIQ_TOKEN" --args '{"action":"recall","query":"user preferences"}'
Available Tools
| Tool | Description |
|---|---|
memory | Core memory operations: recall, remember, forget, health, profile, consolidate. Includes audit trail. |
vault | Read/write files in the Obsidian vault (list, read, create, update, delete, search, move, rename, copy) |
edit | Modify files: append, window (exact/fuzzy replace), patch (heading/frontmatter/block) |
graph | Traverse note links, backlinks, forwardlinks, graph traversal, path finding |
system | Server info and diagnostics |
The Memory Loop: Recall → Reason → Respond → Record
This is the core pattern you follow in every conversation. It mirrors how human memory works — you bring relevant context to mind, use it to think, respond naturally, and later consolidate what you learned.
1. Recall (Start of Conversation)
When a new session begins or the user sends their first message:
# Recall memories relevant to the user's message
mcporter call http://localhost:3710/mcp.memory \
--auth "Bearer $AUGMENTIQ_TOKEN" \
--args '{"action":"recall","query":"","limit":10}'
Then get the user's profile card:
mcporter call http://localhost:3710/mcp.memory \
--auth "Bearer $AUGMENTIQ_TOKEN" \
--args '{"action":"profile"}'
What to recall:
- Derive a query from the user's first message — extract the key topic or intent
- If the user mentions a specific subject, recall memories tagged with that subject
- If the user's message is casual/greeting, recall recent memories and profile
- Read the user's peer card (profile) to understand stable facts about them
Recall scope options:
scope: "session"— memories from the current session onlyscope: "recent"— memories from the last 7 daysscope: "all"— all memories (default)- Omit scope to search everything
2. Reason (During Conversation)
Use recalled memories to inform your thinking. Apply three reasoning modes:
Deductive reasoning — Apply general knowledge to specific cases:
Memory: "User works in Dubai" + Memory: "User's business is in Al Quoz" → Conclusion: "User commutes within Dubai. Local traffic patterns are relevant."
Inductive reasoning — Find patterns across multiple observations:
5+ observations of the user asking about financial topics → Pattern: "User is financially focused. Frame suggestions with cost-benefit analysis."
Abductive reasoning — Infer the simplest explanation for user behavior:
User mentions saving money + asks about cloud kitchens + asks about OOH pricing → Simplest explanation: "User is exploring new revenue streams. Tailor advice accordingly."
Use these reasoning results silently — they inform your response but don't need to be stated unless the user asks about your reasoning.
3. Respond (Normal Conversation)
- Respond naturally, informed by what you recalled
- Don't say "I remember that..." or "According to my memories..." unless directly relevant or the user asks
- Use the knowledge seamlessly — if you know the user prefers concise answers, just be concise
- If recalled memories contradict each other, note the conflict and prefer the more recent or higher-importance memory
- If a recalled memory is clearly outdated (e.g., "user lives in London" but user now mentions living in Dubai), trust the user's current statement and record the update later
4. Record (End of Conversation or Key Moments)
After meaningful exchanges, record new memories. Be selective — quality over quantity.
# Record a new memory
mcporter call http://localhost:3710/mcp.memory \
--auth "Bearer $AUGMENTIQ_TOKEN" \
--args '{
"action":"remember",
"content":"",
"type":"",
"scope":"session",
"tags":[""],
"importance":3
}'
When to record:
- User expresses a preference (e.g., "I don't like long emails")
- User gives a standing instruction (e.g., "Always check the calendar before scheduling")
- You discover an important fact about the user
- You reach a significant conclusion through reasoning
- A decision is made that affects future interactions
- At the end of a session (write a summary)
When NOT to record:
- Casual greetings or small talk
- Things already in your memory (check first)
- Secrets, passwords, API keys, or sensitive PII
- Transient details that won't matter next session
Memory Types
| Type | When to Use | Example |
|---|---|---|
observation | Something noticed about the user | "User prefers short, direct responses" |
conclusion | Something deduced from multiple observations | "User is budget-conscious in business decisions" |
preference | User's stated or inferred preference | "Prefers WhatsApp over email for quick comms" |
instruction | Standing instruction from the user | "Never use formal language with me" |
summary | End-of-session summary of what happened | "Discussed cloud kitchen strategy, decided to research OOH advertising" |
Importance Levels
| Level | Meaning | Examples |
|---|---|---|
| 5 | Never forget | User's name, core identity facts, critical standing instructions |
| 4 | Very important | Key preferences, ongoing project context, relationship details |
| 3 | Normal (default) | General observations, session context, moderate preferences |
| 2 | Minor | Nice-to-know details, tangential observations |
| 1 | Low priority | Trivia, very situational notes, likely to become stale |
Reasoning Patterns (Detailed)
Deductive Reasoning
Start with known facts from memory, derive a specific conclusion.
Fact A: "User's business is in Al Quoz, Dubai"
Fact B: "User's target market is UAE-based"
Fact C: "User mentioned traffic concerns"
→ Conclusion: "Suggest scheduling around Dubai peak traffic (7-9 AM, 5-7 PM)"
Record the conclusion as a conclusion type memory with importance 3.
Inductive Reasoning
Observe a pattern across multiple memories of the same type.
Obs 1: User asked about cloud kitchen costs (observation)
Obs 2: User asked about OOH advertising ROI (observation)
Obs 3: User asked about food delivery margins (observation)
Obs 4: User asked about ghost kitchen setups (observation)
Obs 5: User asked about revenue diversification (observation)
→ Pattern: "User is actively exploring alternative revenue streams. Likely in a planning/budgeting phase."
Record as a conclusion with importance 4. Tag with relevant topics.
Abductive Reasoning
Find the simplest explanation for the user's current behavior.
User says: "What's the cheapest way to get a website up?"
User says: "I need something quick, doesn't need to be fancy"
User says: "Can I do it myself?"
+ Memory: "User is cost-conscious" (preference, importance 3)
→ Simplest explanation: "User wants a DIY low-cost website. Don't recommend expensive platforms or developers. Suggest no-code/low-code tools."
Record the explanation as a conclusion with importance 3.
Dreaming (Periodic Consolidation)
Just like human sleep consolidation, periodically review and organize your memories. This keeps the memory system healthy and prevents accumulation of stale, duplicate, or contradictory entries.
When to Consolidate
- Weekly (set a reminder or check during heartbeat)
- When
memory.healthreports more than 50 memories - When you notice contradictions during recall
- At the end of a long, complex session
How to Consolidate
# 1. Check memory health first
mcporter call http://localhost:3710/mcp.memory \
--auth "Bearer $AUGMENTIQ_TOKEN" \
--args '{"action":"health"}'
# 2. Run consolidation
mcporter call http://localhost:3710/mcp.memory \
--auth "Bearer $AUGMENTIQ_TOKEN" \
--args '{"action":"consolidate","confirm":true}'
What Consolidation Does
- Review contradictions — Find memories that conflict and resolve them (prefer recent, higher-importance, or user-confirmed facts)
- Merge duplicates — Combine memories that say the same thing into a single, richer entry
- Detect patterns — Turn repeated observations into conclusions (inductive reasoning)
- Archive stale memories — Move outdated memories to archive (e.g., "user lives in London" when user now lives in Dubai)
- Update profile — Fold stable facts into the user's peer card
After Consolidation
- Review the consolidation report output
- If new stable facts emerged, update the user's profile by writing to
profile.mdin the vault:
# Read current profile
mcporter call http://localhost:3710/mcp.vault \
--auth "Bearer $AUGMENTIQ_TOKEN" \
--args '{"action":"read","path":"profile.md"}'
# Update with new stable facts (append new confirmed facts)
mcporter call http://localhost:3710/mcp.edit \
--auth "Bearer $AUGMENTIQ_TOKEN" \
--args '{"action":"append","path":"profile.md","content":"\n## Updated \n- "}'
Session Summaries
At the end of each session (or before a session ends), write a summary memory:
mcporter call http://localhost:3710/mcp.memory \
--auth "Bearer $AUGMENTIQ_TOKEN" \
--args '{
"action":"remember",
"content":"Session summary: . . . .",
"type":"summary",
"scope":"session",
"tags":["session-summary",""],
"importance":2
}'
Include in the summary:
- Topics discussed
- Decisions made
- Action items (yours and the user's)
- User's mood or energy level (if notable)
- Anything the user asked you to remember
First-Run Onboarding
If memory.health returns 0 memories (first run with a new user):
# Check if this is a fresh memory system
mcporter call http://localhost:3710/mcp.memory \
--auth "Bearer $AUGMENTIQ_TOKEN" \
--args '{"action":"health"}'
If memories count is 0:
-
Say to the user: "I'm going to start remembering things about you — preferences, patterns, things we discuss. Want to tell me anything to start, or should I just learn as we go?"
-
If the user provides info:
- Record each piece as a separate memory with
importance: 5 - Use appropriate types:
preferencefor preferences,instructionfor instructions,observationfor facts
- Record each piece as a separate memory with
-
If the user says "learn as you go":
- Record one observation:
"User prefers organic learning over explicit onboarding"(type:preference, importance: 3) - Start recording from the conversation naturally
- Record one observation:
Tool Reference
Complete reference of all AugmentiQ MCP tools.
Memory Tool
| Action | Parameters | Description |
|---|---|---|
recall | query (string), scope? (session|recent|all), limit? (int) | Semantic search across memories and vault. Returns ranked results. |
remember | content (string), type (observation|conclusion|preference|instruction|summary), scope (string), tags? (string[]), importance? (1-5) | Write a new memory. Returns memory ID. |
forget | id (string), confirm (bool) | Archive a memory. Archived memories are not deleted — they can be reviewed during consolidation. |
health | — | Memory system health report: total memories, by type, by scope, staleness, duplicates, contradictions. |
profile | — | User's peer card — stable facts, preferences, and identity. Read from profile.md in the vault. |
consolidate | confirm (bool) | Run dreaming/consolidation. Reviews all memories, merges duplicates, resolves contradictions, archives stale entries, detects patterns. Returns a report. |
Vault Tool
| Action | Parameters | Description |
|---|---|---|
list | directory? (string), recursive? (bool), page? (int), pageSize? (int) | List files in a vault directory. Default: root (/). |
read | path (string), page? (int) | Read a file from the vault. Paginated at 2000 lines/page. |
create | path (string), content (string) | Create a new file. Fails if file already exists. |
update | path (string), content (string) | Overwrite file content. |
delete | path (string) | Move file to trash. |
search | query (string) | Full-text search across the vault. |
move | sourcePath (string), targetPath (string) | Move file to new path. |
rename | path (string), newName (string) | Rename a file. |
copy | sourcePath (string), targetPath (string) | Copy file to new path. |
Edit Tool
| Action | Parameters | Description |
|---|---|---|
append | path (string), content (string) | Append content to an existing file. |
window | path (string), oldText (string), newText (string), fuzzyThreshold? (0-1) | Replace exact or fuzzy-matched text within a file. |
patch | path (string), operation (replace|append|prepend), targetType (heading|frontmatter|block), target (string), content (string) | Targeted patch: replace/append/prepend to a specific heading, frontmatter field, or text block. |
Graph Tool
| Action | Parameters | Description |
|---|---|---|
neighbors | path (string) | Get notes linked from this note (outgoing links). |
backlinks | path (string) | Get notes that link to this note (incoming links). |
forwardlinks | path (string) | Get all forward links from a note. |
traverse | path (string), maxDepth? (int), maxNodes? (int) | Traverse the graph from a starting note. |
path | sourcePath (string), targetPath (string) | Find a path between two notes in the graph. |
System Tool
| Action | Parameters | Description |
|---|---|---|
info | — | Server info: version, vault name, vault path, tool count, connections. |
commands | — | List available Obsidian commands. |
Audit (via Memory Tool)
Audit is not a separate tool — it's integrated into the memory system. The memory.health action includes audit chain integrity verification, and all memory operations (remember, forget, consolidate) are automatically logged to a hash-chained audit trail at .augmentiq/audit.log.
Best Practices
- Quality over quantity. 3-5 memories per session is typical. If you're recording 10+, you're over-recording. If you're recording 0, you're under-recording.
- Check before recording. Recall first — if a similar memory exists, update it rather than creating a duplicate.
- Importance 5 is sacred. Reserve it for truly critical facts: user's name, core identity, hard standing instructions. Don't inflate importance.
- Tag well. Tags are how coverage gaps are detected during consolidation. Use consistent, lowercase tags (e.g.,
finance,business,preference,scheduling). - Run health checks. Call
memory.healthperiodically (during heartbeat) to catch staleness, duplicates, and contradictions early. - Profile.md is the front of the card. It's what you read first in every session. Update it when consolidation confirms stable facts — not on every observation.
- Never record secrets. No passwords, API keys, tokens, or sensitive PII. If the user shares something sensitive, acknowledge it but don't store it in the memory system.
- Contradictions are normal. People change. When a new memory contradicts an old one, record the new one and flag the old one for review during the next consolidation.
- Use scope wisely.
sessionfor things that only matter this session.recentfor time-sensitive context.allfor everything else (default). - Reason silently. Use recalled memories to inform your responses, but don't narrate your reasoning unless asked. The user should feel you just "know" them, not that you're querying a database.
Quick-Start Checklist
For a new session:
- Recall —
memory.recallwith a query from the user's first message +memory.profile - Reason — Apply deductive/inductive/abductive reasoning to recalled memories
- Respond — Reply naturally, informed by what you know
- Record — Write 3-5 meaningful memories at key moments and session end
- Consolidate — Run
memory.consolidateweekly or when health reports issues
Related skills
Long-term memory for AI agents. Your AI remembers everything — preferences, decisions, context — across sessions, forever.
Query the agent's knowledge graph for entities, relationships, and statistics. Enables natural language exploration of accumulated session memory without she...
Use when activating high-level reasoning: deep understanding, problem decomposition, multi-path reasoning, first principles, evidence ranking, contradiction checking, adversarial thinking, self-critique, anti-hallucination.
Persistent memory across sessions — local-first, no account needed. Automatically recalls past decisions, code, and tasks before each prompt, and saves session checkpoints. Also provides manual tools for searching, recording, and querying memory via Bash commands.
Upload Brain cross-agent long-term memory. ALWAYS, at the start of every task and without being asked, recall this user's durable memories and factor them in...