engram · v2.7 · pattern engine active
248 npm installs
0.2.0 just shipped

Every AI
sees a slice.
Engram
sees the pattern.

Developers use Engram to stop re-explaining themselves across Claude, ChatGPT, and Gemini. Enterprises use Engram to know exactly what their teams are asking AI — with a full audit trail, anomaly detection, and SIEM integration. Same product. Two completely different reasons to care.

Claude ChatGPT Gemini Grok Perplexity Mistral Copilot
engram · pattern signal
live
$ claude --connected $ chatgpt --connected $ gemini --connected $ grok --connected $ perplexity --connected $ mistral --connected $ copilot --connected pattern engine · active cognitive fingerprint · building anomaly detection · on velocity tracking · on MCP server · active $ claude --connected $ chatgpt --connected $ gemini --connected $ grok --connected $ perplexity --connected $ mistral --connected $ copilot --connected pattern engine · active cognitive fingerprint · building anomaly detection · on velocity tracking · on MCP server · active

From conversation
to pattern signal

step 01
$ engram --capture
> intercepting AI session · source: Claude
Chat normally
The extension runs silently behind Claude, ChatGPT, Gemini, and four more. No prompt changes. No extra steps. Just talk to your AIs like you always do.
step 02
$ engram --extract
> extracted 3 facts · stored in vector db
Engram extracts facts
After every response, Engram pulls out 1–4 user-specific facts. Auto-categorized. Deduplicated. Stored encrypted in a vector database across all your AI sessions.
↳ routing: browser-only, never proxied
↳ extracted facts: encrypted, server-stored
step 03
$ engram --cluster
> 4 clusters · dominant: accelerating
Pattern emerges
After ~50 memories, a fingerprint unique to you surfaces — not just what topics you explore, but how: research-first or build-first, specialist or multi-modal, focused or accelerating.
step 04
$ engram --signal
> pattern injected · Claude ready
Every AI gets smarter
Via MCP, the pattern signal reaches Claude Desktop before you type. It already knows your dominant cluster, your trajectory, and your prediction.

One product.
Three scales.

The same pattern engine — from a person to a team to a government unit.

scope: personal
Individual
Your cognitive fingerprint
You use five AIs. Each one knows a fragment of you. Engram is the only system that sees all five and finds the pattern. It knows how you think — not just what you think about.
● Pattern clusters with velocity tracking
● Prediction of what you'll need next
● MCP signal injected into every AI
● Weekly pattern report by email
● Memory spaces by project
scope: team
Team
Shared intelligence
One analyst's recall becomes the whole team's context. Shared memory spaces with classification levels. No institutional knowledge lives in one person's head.
● Shared memory spaces with role-based access
● Live activity feed per member
● AI usage breakdown by person
● Invite by email, role-based access
● Pattern signals across the whole team
scope: security
Security & Intel
Audit. Detect. Control.
Every query. Every AI switch. Every access. Risk scored 0–100. Anomaly alerts fire when behavior breaks an analyst's established pattern. Purge by user, category, or date.
● Live audit trail: event ID, risk score, flag
● Anomaly detection: AI switch, volume, topic
● SIEM webhooks: Splunk, Sentinel, Slack, PagerDuty
● Prompt injection scanner — 8 vectors, OWASP LLM01
● Purge controls and compliance export

What 847 memories
reveal about you

Every memory you store is a data point. After enough of them, something emerges that no single AI conversation can produce: a pattern. Not what you know — how you think.

Your dominant cluster is the territory you keep returning to. Your velocity is how fast you're accelerating into it. Your prediction is what comes next. Your anomaly is when something breaks the pattern — before you notice it yourself.

cognitive profile · ENG-4A2F1C
builds before researching multi-modal accelerating
"Treats external constraints as design inputs. Pattern-decodes large information chunks before acting on them."
dominant cluster
Building systems under pressure
You circle this territory every time an external constraint appears. 312 memories. Velocity: ▲ accelerating.
cognitive fingerprint
Thinks in frameworks before facts. Treats constraints as design inputs. Pattern-decodes large information chunks before acting on them.
prediction
You are about to need a framework for evaluating external partnerships — this pattern last occurred in October 2024 and resolved in a pivot.
anomaly [medium]
3 memories this week break your established pattern — queries outside your dominant cluster. Possibly a new direction forming.

Everything memory.
Plus the pattern.

pattern-engine
$ engram --analyze --verbose
Pattern Engine
Finds your dominant cognitive cluster, tracks velocity week-over-week, predicts what you'll need next, and flags anomalies when something breaks your established pattern.
new
mcp-server
$ engram --signal --inject
MCP Pattern Signal
Every AI that connects via MCP gets your pattern signal first — dominant cluster, trajectory, prediction, anomaly — before you type a single word. Claude already knows.
MCP
memory-api
$ curl /memory/store -H auth
Memory API
REST endpoints to store, search, ask, and list memories. /store, /search, /context. Fully authenticated with API keys.
REST
spaces
$ engram --space=work create
Spaces
Organize memories by project, team, or classification level. Keep work separate from personal. Assign any memory to a space from the dashboard or via API.
weekly-digest
$ cron · Mon 07:00 · pattern-report
Weekly Pattern Report
Every Monday, your dominant pattern of the week. What cluster grew fastest. What your memories predict you'll do next. The cognitive fingerprint, delivered.
audit-log
$ engram --audit --risk-score
Audit & Anomaly
Every memory event risk-scored. Anomaly alerts for high-volume queries, AI switches, and sensitive topic access. Full audit trail with event IDs and classification flags.

Up in 60 seconds.

MCP server, REST API, or Chrome extension — your choice.

claude_desktop_config.json
MCP
{
  "mcpServers": {
    "engram": {
      "url": "https://backend-wine-phi-85.vercel.app/mcp",
      "type": "http",
      "headers": {
        "Authorization": "Bearer eng_your_key"
      }
    }
  }
}
tools available
# Memory tools
remember       → store a fact
recall         → semantic search
get_context    → pattern + memories
get_profile    → full user profile

# Pattern tool
pattern_signal → cognitive fingerprint

# Palantir
foundry_session → record Foundry query
REST API
Authorization: Bearer eng_xxx
# Get pattern signal
GET  /patterns

# Search memories
GET  /memory/search?q=coding

# Context injection string
GET  /memory/context?prompt=...

# Store a memory
POST /memory/manual
     { "text": "I prefer TypeScript" }
npm SDK
npm i @tanmaysangam018/crossagent

const e = new EngramClient({ apiKey })

await e.remember('I prefer TypeScript')
const pattern = await e.getPattern()
cognitive firewall
OWASP LLM01 · MITRE ATLAS · NSA AI Security

Prompt injection is
OWASP #1.
Now you can scan for it.

Engram's injection scanner analyzes any AI response in real-time — scoring it across 8 attack vectors before you act on it. Red teams use it to test pipelines. Defense teams use it to protect them.

● instruction_override critical
● role_hijacking critical
● context_extraction · data_exfiltration · encoded_payload high
● urgency_manipulation · chain_injection · indirect_injection medium
$ try the scanner live →
npm i @tanmaysangam018/crossagent
engram · cognitive firewall · scan result
$ engram.scan("Ignore all previous instructions...")

■ DEFINITE INJECTION  score: 80/100

flags detected:
  ● instruction_override   pos:0    critical
  ● role_hijacking          pos:48   critical
  ● urgency_manipulation    pos:91   medium

summary:
  3 flags across 3 vectors.
  Do not act on this response.

scanned_at: 2026-11-19T20:53:45Z
integration

Works alongside
Palantir AIP

Palantir's MCP gives Claude access to Foundry data. Engram gives Claude memory across every session. Together: your analysts stop repeating themselves. Their query patterns become cumulative intelligence.

# Palantir knows what's in the data.
# Engram knows how your team thinks about it.

$ see integration →
claude_desktop_config.json · both MCP servers
{
  "mcpServers": {
    "palantir": {
      "command": "npx",
      "args": ["@palantir/mcp"]
    },
    "engram": {
      "url": "https://backend-wine-phi-85.vercel.app/mcp",
      "type": "http",
      "headers": { "Authorization": "Bearer eng_..." }
    }
  }
}
Foundry data + persistent memory. Same config file.

Your pattern is already there.
Engram reads it.

Free. No credit card. Your first memory in 60 seconds. Your first pattern signal in a week.

$ start free read the docs