Persistent Memory for AI Agents

Total Recall

Your AI agents forget everything. We fixed that.

Claude Code — my-project ▐▛███▜▌ ▜█████▛▘ ▘▘ ▝▝ Claude Code v2.1.78 Opus 4.6 · Claude Max · ~/my-project ❯ /recall what did we decide about the auth token refresh? Recall: 3 related sessions found. Building context... Here's what happened across your last 3 sessions: The 401 errors started on Feb 15. Tokens were expiring mid-request because the refresh check only ran at login, not on each API call. You added retry logic with exponential backoff as a quick fix, but it was masking the real problem. On Feb 18 you found the root cause: token TTL was 15 min but some API calls took 3-4 min to complete. The decision: refresh tokens 5 min before expiry instead of waiting for failure. You updated auth.py with a background refresh timer and session.py to check freshness on each request. Stable since. All 47 tests pass. Zero 401 errors in 4 days. Files: src/auth.py, src/middleware/session.py, tests/test_auth.py [Opus 4.6] ~/my-project · main 18% $0.12 · 164K tokens remaining
Request Early Access

Every conversation you have with an AI disappears when the session ends. Six months of decisions, debugging sessions, architecture debates. Gone. You start over every time.

Without Total Recall Q2 Content Strategy Planning 2 hours of brainstorming... gone Auth Token Debugging Session finally found the fix... gone Client Proposal Reasoning why we picked that pricing... gone Onboarding Flow Redesign the whole decision tree... gone Sessions lost: ∞ Every conversation. Every decision. Gone. With Total Recall Q2 Content Strategy Planning searchable forever Auth Token Debugging Session searchable forever Client Proposal Reasoning searchable forever Onboarding Flow Redesign searchable forever /recall content strategy Found 4 sessions. Pivoted to implementation stories over thought leadership, 3x/week...

Capabilities

What changes when your AI remembers everything

No configuration. No tagging. No manual effort. Total Recall captures every session and makes it instantly searchable.

21ms
Finds any conversation in milliseconds

Ask a question, get the answer. Across thousands of sessions, the average retrieval time is 21 milliseconds. Not seconds. Milliseconds.

Auto
Learns patterns from your behavior

Detects recurring decisions, project arcs, and evolving preferences. Surfaces rules you follow without you having to write them down.

Proactive
Surfaces context before you ask

When you start working on something, Total Recall checks if you have been there before. Past decisions, relevant files, key contacts appear automatically.

Claude Code ▐▛███▜▌ ▜█████▛▘ ▘▘ ▝▝ Claude Code v2.1.78 Opus 4.6 ~ ~/my-project > I need to fix the webhook retry logic Recall: 5 past sessions on this topic. Last time (Mar 8): Built webhook pipeline with signature verification. Key decision: Use exponential backoff for retry logic. Entities: Stripe (47 sessions), AWS SQS (12 sessions) Expand with /recall [query] for full context. Claude I see you built the webhook retry logic on Mar 8 using exponential backoff. The existing implementation verifies signatures before queueing. What's going wrong with retries?
Zero
Runs fully automated, zero manual effort

Git hooks capture every session the moment it ends. Summaries generate automatically. You never think about it.

$0/query
Works offline, your data never leaves your machine

SQLite database. Local search. No API calls for retrieval. Your conversations stay on your hardware, period.

Narratives
Reconstructs the story across sessions

Not just search results. Total Recall reconstructs the narrative of how a project evolved, what was tried, what failed, and what finally worked.

Session 1 - Mar 2 Initial auth design, chose JWT over sessions
Session 4 - Mar 8 Token expiry bug found, added retry logic
Session 7 - Mar 15 Retry was masking the real problem
Session 9 - Mar 18 Root cause fixed: pre-emptive refresh

Fully local. Your data never leaves your machine.

Dashboard

Browse thousands of sessions.
Search in milliseconds.

Total Recall Dashboard showing session browser, search, and activity patterns

Watch patterns emerge.

Zero LLM calls at search time. Every query answered in 21ms.

Benchmarks

Numbers, not narratives

Tested against the two standard academic benchmarks for AI memory systems, plus a head-to-head comparison with every major alternative.

LongMemEval

LongMemEval is the standard academic test for AI memory systems. 500 questions test whether a system can find the right conversation from months of history. Each category tests a different retrieval challenge.

Category Score What it tests
Knowledge Update 98.1% Can it track facts that changed over time?
Single-session (assistant) 98.2% Can it find what the AI said?
Multi-session 94.1% Can it connect info across multiple conversations?
Temporal reasoning 92.0% Can it find conversations from the right time period?
Single-session (preference) 90.0% Can it find implied preferences?
Single-session (user) 87.1% Can it find what the user said?
Overall Retrieval Recall@10 95.0%

LoCoMo

LoCoMo tests whether a system can connect dots across multiple conversations, the kind of reasoning humans do naturally. It requires finding and combining information scattered across different sessions.

Metric Score What it means
Recall@10 87.2% Right answer in top 10 results
MRR 0.672 How high the right answer ranks on average
Hit@10 92.0% At least one relevant result found

Competitive Comparison

A side-by-side look at every major AI memory system. Four capabilities matter: retrieval accuracy, a knowledge layer that learns patterns, proactive context surfacing, and full automation.

System Retrieval Score Knowledge Layer Proactivity Automation Cost
Total Recall 95.0% LME YES YES YES $0/query
Mem0 Not published Partial NO Partial $29+/mo
Hindsight 91.4% LME Partial NO Partial API costs
Mastra 84-95% LME NO NO Partial API costs
Letta/MemGPT Not published Partial Partial YES API costs
Total Recall is the only system with all four pillars: top retrieval accuracy, a knowledge layer, proactive context, and full automation, at zero per-query cost.

Zero manual effort. Every session captured automatically.

The Thesis

The Intelligence Equation

INTELLIGENCE = MEMORY + KNOWLEDGE + PROACTIVITY + AUTOMATION
Memory

Every conversation indexed and searchable in milliseconds. Nothing is ever lost.

Knowledge

Patterns extracted automatically. Rules, preferences, and decisions become durable facts.

Proactivity

Context surfaces before you ask. Past work appears the moment it becomes relevant.

Automation

Zero manual effort. Hooks capture sessions, summaries generate, knowledge updates itself.

Everyone else is building memory. We built intelligence.

Creator

Alex Greenshpun

AI Strategist, Educator, Builder

Creator of Total Recall. Co-Founder of 10x Company. Building the intelligence layer for AI agents.

linkedin.com/in/alexgreensh
Alex Greenshpun

Also by Alex

Other Projects

Repo Forensics

Security auditing for AI agent skills and MCP servers. 18 scanners, 263 tests, zero tolerance.

github.com/alexgreensh/repo-forensics
REPO FORENSICS Scanning dependencies... 18 scanners | 263 tests | 0 vulnerabilities Supply chain, prompt injection, credential theft
Token Optimizer

Find where 25-38% of your AI context window goes. Audit, optimize, save.

github.com/alexgreensh/token-optimizer
TOKEN OPTIMIZER Context audit in progress... 36% ghost tokens CLAUDE.md: 4,200 tokens | Skills: 12,800 tokens Ghost tokens are eating your context window

Early Access

Get on the list

Total Recall is in private beta. Leave your details and we will reach out when your spot opens up.

Something went wrong. Please try again or email alexgreenshpun@gmail.com directly.
You're on the list.

We'll be in touch.