Real Dakera API calls in your browser. Chat sessions, hybrid search tuning, entity extraction, multi-agent sharing, knowledge graphs, guided scenarios, and LLM comparison.
What makes Dakera different: Every stored memory carries an importance score (0–1) that controls exponential decay — critical facts persist indefinitely while ephemeral context fades naturally. Recall combines vector similarity + importance weight + temporal recency into a single smart score, achieving 88.2% accuracy on the LoCoMo long-context memory benchmark.
88.2% LoCoMo accuracy·<10ms recall latency·9 live modes·5 languages·Chat sessions
Live sandbox — this code runs against our real Dakera API. No account, no setup. Click ▶ Run to execute it now.
playground.py
30 req/min · 50 mem/session · ⌘⏎ to run
0.80
output.jsonidle
// Press ▶ Run to execute…
✨ Live Visualization
// Run a scenario to see the visualization
Compare Recall Strategies
Same query, three depth levels. See how top_k shapes what your agent retrieves — breadth vs. precision.
Broad Recall
top_k=10 · all results ranked
Returns up to 10 memories ranked by smart score. Best for discovery — you see everything the agent knows about this topic, including loosely related facts. Use when you want maximum context.
discovery mode
▶ Click Compare to run
Balanced
top_k=5 · top half of results
Returns the top 5 most relevant memories. Filters out peripheral context while keeping all meaningful signals. The default sweet spot for most agent use cases — not too noisy, not too narrow.
default mode
▶ Click Compare to run
Precise
top_k=3 · highest relevance only
Returns only the 3 highest-scoring memories. Maximum signal-to-noise ratio. Use when your prompt window is tight or you need the agent to act on only its strongest beliefs — ideal for critical decisions.
precision mode
▶ Click Compare to run
Smart Score = vector similarity × importance weight × temporal recency factor — a single 0–100% relevance signal combining semantic match, how critical the memory is, and how recently it was stored or accessed.
Free-form API Explorer
Call any Dakera endpoint directly — craft your own request, see the raw response.
RequestPOST
Request Body (JSON)
Responseidle
// Response will appear here…
AI With Memory vs. Without
Ask the same question to an LLM with and without Dakera memory. See exactly how persistent context changes the answer — and which memories made the difference.
Powered by OpenRouter free models — Llama 4 Maverick or Gemma 3. Real LLM inference, real memory retrieval.
Try a preset scenario:
1Seed memories
2Recall context
3LLM inference
4Compare
🚫 Without Memory
LLM has no context about your data
Select a preset or type a question, then click Compare…
Generic response — no domain knowledge
✅ With Dakera Memory
LLM receives recalled memories as context
Select a preset or type a question, then click Compare…
Precise, contextual response — powered by recalled memories
🔍 Memories injected as context:
ChatMemorySession — Live Demo
A real conversation with persistent memory. Each message you send is stored via Dakera. Subsequent messages recall prior context — the agent remembers.
Session not started
Raw API Responses
Hybrid Search Tuner
Adjust the balance between BM25 keyword search and vector semantic search. See how results change in real-time.
BM25 (keyword)Vector (semantic)
vector_weight: 0.50
Seed memories, then search to see results
Entity Extraction — Live
Paste or type text. Dakera extracts named entities (people, orgs, locations, dates) with highlighted spans.
Type text above and click Extract
Multi-Agent Memory Sharing
Agent A stores memories. Agent B recalls them. Zero re-introduction — instant cross-agent context via Dakera.
Agent A — Onboarding Bot
0.85
➡
Agent B — Support Bot
Knowledge Graph Explorer
Store memories, then explore the auto-built knowledge graph. Click nodes to traverse relationships.
Seed memories and build graph to explore
Node Details
Ready to Add Memory to Your Agent?
Self-hosted, open-core, one binary. From zero to recalled memories in under 5 minutes.
DAKERA