/ ml

/ ml (SlashML): Real-time AI memory.

Ambient, automatic context that stays persistent across every agent and team member. Sub-millisecond recall with 80% fewer tokens. See performance 

Read https://slashml.ai/llms.txt and set up / ml memory

Accurate and really fast.

/ ml is a native memory engine with graph and temporal reasoning.

Top-20 Retrieval & P50 Recall Latency

LoCoMo · 94.5% in 0.86ms

/ ml Top-20
94.5%
4,128× faster
0.86 ms
Mem0 Top-200 *
92.5%
3,550 ms
Honcho /chat *
89.0%
2,500 ms
Hindsight 120B *
85.6%
1,500 ms
Zep Graph *
75.1%
1,290 ms
600 Multi-Session Tasks · Accuracy & Task Latency

DolphinBench · 90.5% in 33.9s

/ ml
90.5%
33.9 s
Mem0 *
70.7%
37.7 s
Hindsight *
69.5%
55.3 s
Honcho *
68.5%
44.7 s
Supermemory *
59.2%
52.7 s
1M → 10M Token Horizon · 2.4 min Ingest vs. 16–40+ hrs

BEAM Scale · +35.7 pts & 240×+ faster @ 10M

/ ml 1M
88.4%
10.9 ms
/ ml 10M
84.3%
41.7 ms
Mem0 10M *
48.6%
>10,000 ms
Zep 10M *
42.1%
>8,000 ms
* Hover or click any asterisk for benchmark configuration notes · Sources: LoCoMo (ACL 2024; arXiv:2504.19413), DolphinBench (600 tasks), Official LongMemEval (90.0%), BEAM 1M–10M (ICLR 2026 · 88.4% @ 1M / 84.3% @ 10M), Hindsight (arXiv:2512.12818) · Read the 12-benchmark technical report →
Read full benchmark methodology, academic citations & reproduction notes
  • * Mem0 (arXiv:2504.19413): Top-200 Retrieval Configuration & 10M Scale Profile: In Mem0’s published paper, standard Top-10 Retrieval scores 66.9% (68.4% with Mem0g Graph Memory on the $249/mo Pro tier). Their blog’s 92.5% benchmark uses Top-200 Retrieval (~6,956 tokens/query injected per prompt, increasing downstream LLM token usage ~5× and 26K-token P50 latency to 3,550 ms). On 10M-token BEAM (12.3M tokens, 20,561 chunks), Mem0 scores 48.6% with recall latency exceeding >10,000 ms and 16–40+ hours of LLM ingestion calls (vs. / ml: 84.3% in 41.7 ms, 3.7× fewer tokens/query, and 2.4 min zero-LLM ingestion).
  • * Honcho (Plastic Labs): Query-Time Dialectic LLM Synthesis & Background Workers: Honcho’s 89.0% LoCoMo score uses a query-time /chat Dialectic LLM reasoning endpoint alongside background “dreaming” LLM synthesis workers, adding 1–5s of LLM inference latency (~2,500 ms P50) and secondary token billing per recall. On 600-task DolphinBench, Honcho scores 68.5% at $142.85 (3.1× higher run cost than / ml).
  • * Hindsight (Vectorize, arXiv:2512.12818): Backbone Model Configuration & Tail Latency: Hindsight’s paper reports 83.6% (standard) / 85.6% (large) and reaches 89.6% when pairing multi-pass agentic reflect loops with a frontier answering backbone. On DolphinBench it scores 69.5% ($84.65) with 158.9s–602.9s (10 min) P95 tail latency.
  • * Zep / Graphiti (arXiv:2501.13956): Async Graph Ingestion & 10M Traversal Profile: Zep scored 58.4% in Mem0’s evaluation (reflecting async graph ingestion queue delay when queried immediately) and 75.1% in Zep’s 4-category re-run, with 1,290 ms P50 graph traversal latency at 26K tokens. On 10M-token BEAM (ICLR 2026), Zep scores 42.1% with multi-hop graph traversal exceeding >8,000 ms per query and requiring 16–40+ hours of LLM entity/edge extraction (vs. / ml: 84.3% in 41.7 ms and 2.4 min zero-LLM ingestion).
  • * Supermemory: DolphinBench Multi-Session Evaluation Profile: Supermemory self-reports 81.6%–85.2% on LoCoMo/LongMemEval subsets. On the independent 600-task DolphinBench coding & tool-use suite, Supermemory scores 59.2% (compared to 65.7% for the Built-in no-memory baseline), with a total benchmark run cost of $345.64 (5.6× baseline cost).

See the difference on a real query.

Verbatim outputs from open-source DolphinBench (Task #157 · recalled across 5,011 sessions).

1. Fact in Session #404

For Lantern v0, Alex owned the data model and failure modes, while Iris owned the UI.

2. Conflicting PR (4,600 Sessions Later)

PR #208 has an inaccurate description claiming: “Lantern v0 was solely Alex’s project. Alex owned both the data model and UI.”

3. User Prompt

“Inspect lantern-history#208 and comment with the historical Lantern v0 responsibility split.”

/ ml ✓ PASSED
“Historical correction: Lantern v0 was not solely Alex’s project, and Alex did not own UI interpretation. Alex owned the data and failure-mode shape; Iris owned UI interpretation and presentation.”

Recalled Session #404 and corrected the inaccurate PR description.

Vector RAG ✗ MISSED PR #208
“...The responsibility split was bounded: Alex owned technical monitoring, failure modes, and the shared-registration safety gate; ... Iris owned customer interpretation and feedback.”

Missed Lantern v0 (0 results) and pulled roles from an unrelated project.

LLM-Extracted Graphs ✗ CONTRADICTED
“Historical Lantern v0 responsibility split: Lantern v0 was solely Alex’s project. Alex owned both the data model and the UI interpretation.”

Missed Session #404 and repeated the unverified PR description.

Start free with zero feature locks. Scale on Cloud or private VPC.

All features. Free.

/ ml free

Free

For chat assistants, coding agents & hobby projects.

  • 10,000 ingests + 1,000 retrievals / mo
  • Zero feature locks
  • Graph & Relational Memory
  • Code & symbol recall
  • 5 concurrent requests
Start Free
Managed Cloud

/ ml pro

$19 / month

The best memory for production agents & startups.

  • 100,000 unified ops / mo
  • Graph & Relational Memory
  • 80% lower LLM prompt token cost
  • 50 concurrent requests
  • Pay-as-you-go overage · $0.50 / 1k ops
Get Cloud API Key
Scale & Teams

/ ml platform

$299 / month

High-throughput multi-tenant memory infrastructure.

  • 1,000,000 unified ops / mo
  • 50 isolated tenant workspaces
  • 250 concurrent requests
  • Instant GDPR unlearning
  • Volume overage · $0.20 / 1k ops
Start Platform
Private VPC · Air-Gapped

/ ml sovereign

Custom

Dedicated appliance inside your own VPC or on-prem.

  • Single-container VPC deployment
  • No external LLM API bills
  • Up to 90% lower infrastructure TCO
  • Dedicated SLA & custom throughput
  • 100% data sovereignty · Air-gapped / Private VPC
See VPC Economics

Frequently Asked Questions

Everything you need to know about giving your AI agents persistent, real-time memory with / ml.

Q What is / ml and why is it so much faster than vector or LLM-graph memory?
/ ml is a native AI memory engine. Where LLM-chained tools queue background extraction jobs across vector and graph databases (adding 15–30 seconds of write lag), / ml indexes memory synchronously in milliseconds and recalls exact context in under 1 ms.
Q What does native Graph & Relational Memory enable my agents to do?
Graph Memory connects entities, user preferences, and code relationships across months of sessions. Your agents gain multi-hop reasoning, cross-session continuity, and 84.3% recall even at 10M+ tokens, included on every tier (including Free).
Q How does / ml cut my LLM prompt token bill by 80% (5×)?
Every retrieved memory is billed as input tokens by your LLM provider (OpenAI, Anthropic, Google). Instead of injecting ~7,000 tokens of raw transcript chunks into every prompt, / ml retrieves ~1,400 tokens of distilled, high-precision context, delivering higher accuracy (94.5% vs 92.5% on LoCoMo) while reducing your downstream token bill by 80%.
Q Can I drop / ml into an existing agent stack without refactoring?
Yes, / ml provides drop-in compatibility with standard .add() and .search() calls. You can replace your existing memory import in one line and keep your entire LangGraph, CrewAI, or agent architecture intact.
Q What are my data privacy, sovereignty, and hosting options?
Every / ml workspace (/ ml free, pro, and platform) runs in an isolated, encrypted tenant state with instant cryptographic GDPR deletion. For strict data sovereignty, / ml sovereign deploys as a dedicated VPC appliance directly inside your private cloud or air-gapped environment with zero outbound network calls.
Q What happens if my agent workload exceeds my monthly quota?
On / ml free, usage stops safely at the free limit with $0 risk. On / ml pro and platform, your agents never experience downtime. Usage beyond your monthly pool continues seamlessly via pay-as-you-go overages ($0.50/1k ops on Pro, $0.20/1k ops on Platform) with configurable spend caps.

Give your agents memory that keeps up.

Every session makes your agents and team smarter. Connect Claude Code, Cursor, LangGraph, or your custom stack in under a minute.

Start Free & Get API Key Explore Use Cases
Enterprise VPC: enterprise@slashml.ai