Recallr is the #1 memory layer for conversational AI agents. Persistent, versioned and temporally-aware knowledge, powering chatbots, voice agents, and copilots that never forget.
Evaluated on LongMemEval (oracle) across 500 questions spanning six memory task types.
Single-Session
User
(n=70)
Single-Session
Pref.
(n=30)
Knowledge
Update
(n=78)
Temporal
Reasoning
(n=133)
Multi-
Session
(n=133)
Single-Session
Asst.
(n=56)
Temporal Reasoning Dominance
97.0% vs 50.4% (Mem0 Graph) vs 27.1% (Supermemory)
+46.6pp improvement over the nearest competitor. Recallr knows the difference between when something happened and when you mentioned it. So 'last summer' means last summer, not last Tuesday.
Knowledge Update: 97.4%
97.4% vs 76.9% (Mem0) vs 60.3% (Supermemory)
When facts change, old versions are archived, not erased. Your agent always has the latest truth, and the full history of how it got there.
| Strategy | Min | P25 | P50 | P75 | P95 | Max |
|---|---|---|---|---|---|---|
Recallr (Low-Latency) | 0.234s | 0.265s | 0.299s | 0.338s | 0.408s | 0.750s |
Recallr (Balanced) | 1.032s | 1.132s | 1.198s | 1.286s | 1.575s | 3.548s |
Recallr (Agentic) | 5.125s | 6.194s | 6.997s | 7.765s | 8.619s | 20.095s |
Mem0 | 0.489s | 0.504s | 0.786s | 0.967s | 1.787s | 6.171s |
Mem0 (Graph) | 0.697s | 0.746s | 0.961s | 1.987s | 2.692s | 10.458s |
Supermemory | 0.392s | 0.851s | 1.301s | 1.876s | 3.293s | 4.242s |
voice threshold (0.3s)
chat threshold (1.5s)
<400ms
Real-time voice and chat — answer before the user finishes thinking.
~1.5s
General production workloads — thorough retrieval without perceptible delay.
~8s
Deep reasoning over months of memory — when completeness matters more than speed.
Asynchronous Memory CurationWhile you sleep
Example
You: "I moved to Delhi last month."
Synchronous Context RetrievalFast → Deep · Auto-Recall
Example
You: "What restaurants did I like when I lived in Bengaluru?"
A dead-simple, drop-in replacement for your existing agents. Route your OpenAI, Anthropic, or Gemini calls through Recallr. No SDK changes, no architecture rewrites, no new abstractions.
from openai import OpenAI client = OpenAI( base_url="https://api.openai.com/v1", base_url='https://api.recallrai.com/api/v1/forward/https://api.openai.com/v1', api_key='sk-...', default_headers={ 'X-Recallr-API-Key': 'rai-...', 'X-Recallr-Project-Id': 'your-project-id', 'X-Recallr-Allow-New-User-Creation': 'true', 'X-Recallr-Session-Timeout-Seconds': '600', }) # Memory is automatically injected on every callresponse = client.chat.completions.create( model="gpt-5.5", messages=[{"role": "user", "content": "My name is Mayank"}], extra_headers={ 'X-Recallr-User-Id': 'user-123', 'X-Recallr-Recall-Strategy': 'low_latency', }) print(response.choices[0].message.content)Your existing code works unchanged.
OpenAI, Anthropic, Gemini. One proxy.
Every user gets a persistent profile.
Every domain where context compounds over time. Every agent that should never ask twice.
Healthcare · Longitudinal
Remembers medication reactions, contradictory symptoms, and what the patient said last month versus what they’re saying now. The doctor gets context. The patient gets heard.
Education · Personalization
Knows which concepts clicked, which needed three tries, and how you learn best. Every session builds on the last.
Legal · Case Management
Tracks every witness statement, flags contradictions across depositions months apart, and never loses the thread of a multi-year case. The brief writes itself.
Customer Support · Context
Remembers every ticket, every frustration, every resolution. When a customer returns, the agent already knows their history, preferences, and what was promised last time.
Sales & CRM · Pipeline
Tracks every interaction, objection, and milestone across long sales cycles. Reps get instant recall at every touchpoint — no more re-reading CRM notes before calls.
E-Commerce · Personalization
Remembers size preferences, past purchases, style evolution, and gift recipients. Every recommendation builds on real history, not just last-click behavior.
Chronic Illness · Patient Autonomy
For patients managing chronic conditions, every new doctor visit starts from zero. Recallr gives the AI the full longitudinal picture: symptoms reported six months ago, medications that caused side effects, what "bad days" actually look like for this specific person.
When reported symptoms contradict what was said before, the conflict is surfaced, not silently overwritten. The patient's history becomes a living record, not a forgotten intake form.
Patient Memory Visualization — Oct 2024 → Jan 2025 → Mar 2025
Conflict detected at Oct 2024 · Versions diverged
We give $20 to free accounts every month.
Chat exchanges / session
10
User + assistant pairs per session
Sessions / day
5
Per active user, on average
Simulation window
60 days
Cost projection horizon
Both costs include upstream LLM fees. Naive re-sends full history as context — cost grows quadratically. Recallr uses a fixed memory pipeline per session — cost grows linearly.
Without Recallr
$541.80
full history re-sent every call
With Recallr
$199.80
included in total above
You save
$342.00
63% cheaper than naive
Breakeven
Day 22
Recallr cheaper from here on
Cumulative LLM cost
RAG retrieves static document chunks. Recallr maintains a versioned knowledge graph that evolves over time, with conflict resolution and temporal provenance. It knows what changed, when, and why. RAG just knows what was stored.
Recallr classifies the conflict type (temporal update, correction, preference change, or contradiction) and applies the appropriate resolution strategy. For ambiguous conflicts, it can prompt user clarification rather than silently overwriting.
Yes. The memory layer is model-agnostic. Route your OpenAI, Anthropic, or Gemini calls through Recallr with two lines of code. No SDK changes, no architecture rewrites.
Each memory entity maintains a linked-list version chain. New information creates a new version node with a "supersedes" edge to the previous version, along with dual timestamps (event time and ingestion time) for temporal provenance.
Zero for ingestion. Curation runs asynchronously. For recall: <400ms (Low-Latency), ~1.5s (Balanced), or ~8s (Agentic). Choose the strategy that fits your use case.
Yes. You can query the graph at any point in time using temporal filters. Retrieve what the system knew at a specific date, trace the evolution of a fact, or audit the full version history of any entity.
Stop building agents
that forget.
