Which memory should your AI agent use?
Compare independent test results to see where each memory system works well and where it fails.
Agentic Memory Indexi
MethodologyTap a logo for the tool page.
Agentic Memory Index v0.2 · Sep 2026
Highlights
Accuracy, cost, and speed for every tool, measured the same way.
Agentic Memory Indexi
The overall score: how often each tool finds the right answer.
Costi
Cost / 1,000 questions
Answer speedi
How long an answer takes end to end, same agent model on every tool.
Check your own agent memory
Tell us which memory system you build or use and we’ll show you what the overall score doesn’t tell you.
Which memory tool is right for your agents?
No tool wins everything: match each tool's strengths to the questions your agents actually ask.
Correct Answers by Task Typei
From direct recall to synthesis, each tool has its strengths.
Failure Attributioni
How often each tool fails, and why. Lower is better.
Rankings and retention
Every tool's score, and whether what it stores survives.
Every tool's score, cost, and speed in one table.
| # | Service | Index ↓ | Cost / 1k | Lookups / answerLookups |
|---|---|---|---|---|
| 1 |  Cognee |
97.1 | $256.25 | 1.41 |
| 2 |  Karpathy Wiki |
97.1 | $502.56 | 0.99 |
| 3 |  Mem0 |
95.8 | $252.32 | 1.28 |
| 4 |  Mnemosyne |
95.7 | $302.08 | 1.46 |
| 5 |  Hindsight |
93.7 | $233.19 | 1.20 |
| 6 |  Mitosis Cortex |
92.0 | $270.05 | 1.50 |
| 7 |  gbrain |
90.0 | $453.73 | 0.98 |
| 8 |  Supermemory |
89.9 | $266.47 | 1.39 |
| 9 |  Hyperspell |
89.4 | $249.75 | 1.41 |
| 10 |  Membase |
84.0 | $246.55 | 1.39 |
| 11 |  Stash |
82.0 | $294.53 | 2.45 |
| 12 |  Honcho |
74.2 | $277.84 | 1.43 |
Long-term retentioni
How well early facts survive: probed at five checkpoints across 56 working sessions.
Speed
One-time setup, indexing lag, and per-operation latency.
Onboarding, readiness, and store-session times for every tool.
| Tool | Onboard ↑ | Time to ready | Time to store |
|---|---|---|---|
| Cognee |
7.5 min | 22.4s | 1.4s |
| Hyperspell |
9.6 min | 15.8s | 0.64s |
| Mem0 |
10.6 min | 17.9s | 0.95s |
| Supermemory |
12.8 min | 10.4s | 0.88s |
| Stash |
14.0 min | 15.5s | 1.8s |
| Hindsight |
16.0 min | 160.9s | 2.3s |
| Mnemosyne |
18.1 min | 10.2s | 1.6s |
| Membase |
18.4 min | 18.7s | 1.5s |
| gbrain |
21.3 min | 491.3s | 2.2s |
| Mitosis Cortex |
21.5 min | 62.8s | 0.79s |
| Honcho |
24.2 min | 764.5s | 0.47s |
| Karpathy Wiki |
not measured | not measured | not measured |
Where retrieval time goesi
Per-operation medians, seconds.
Pricing and cost
What the tools charge, and what successful answers actually cost.
What each tool costs to start, from list pricing.
| Tool | Free tier | Starting at | Open source |
|---|---|---|---|
| Cognee |
1M tokens | Pay as you go | Yes |
| Karpathy Wiki |
Always | $0 | Open method |
| Mem0 |
1,000 retrievals/mo | $19/mo | Core is |
| Mnemosyne |
Always | $0 | Yes |
| Hindsight |
Self-host free | Pay as you go | Yes |
| Mitosis Cortex |
7-day trial | $7.99/mo | Closed |
| gbrain |
Always | $0 | Yes |
| Supermemory |
$5 credits/mo | $19/mo | Yes |
| Hyperspell |
Not Listed | Not Listed | Closed |
| Membase |
Limits not listed | $20/mo | Connector is |
| Stash |
10 curator runs/mo | $20/mo | Yes |
| Honcho |
None | Pay as you go | Yes |
Cost Breakdowni
Cost / 1,000 questions