SynapCores
See SynapCores in production. Try one workflow.
Explore live deployment patterns, inspect a working application, then run a related recipe. Before adding another service to your stack, evaluate what SQL, vectors, graphs, and AI can do together.
Production workflows
SynapCores is deployed in the United States, Poland, and Indonesia. These anonymized summaries describe live deployments reported by the SynapCores team.
Production deployment
A call-center deployment uses NLI to check short, relevant agent utterances against a defined rubric as a call progresses. After the call, /v1/systemone evaluates typed review questions over transcript inputs within the model’s context limits, helping surface potential issues for review.
Try the NLI verification patternProduction deployment · private application
A marketing team uses DES, a private application with SynapCores in the backend, to qualify potential customers. Explore a related sample workflow that combines lead scoring, relevant proof, and account relationships.
Explore the sample lead-qualification workflowFirst-party production application
VoiceAgent uses SynapCores inference to guide agents’ behavior in real time. The related recipes let you evaluate supporting voice-agent patterns with your own conversation data and application stack.
Explore a related voice-agent recipeProduction deployment
A credit union uses SynapCores for credit risk scoring and financial fraud workflows that support analyst review. Start your own evaluation with sample transaction anomalies and graph relationships, and inspect the evidence behind a flagged pattern.
Try the sample fraud-review workflowProduction research deployment
A pharmaceutical lab uses SynapCores to support research into new compounds. Explore a related sample graph that connects drugs, targets, and diseases and retrieves mechanism-similar candidates for further investigation.
Inspect the sample research graphTurn a relevant example into your next working result.
Pick the closest workflow and evaluate it on your own infrastructure. The linked recipes use sample data so you can inspect the pattern before adapting it to a real workload.
Install Community EditionFree Community Edition. No signup or credit card.
Inspect what’s already built.
These projects are maintained by SynapCores. Explore the source, follow the data, and use them as starting points for your own evaluation.
First-party reference application
Memory, retrieval, tool routing, and generation use SynapCores. Follow the reference application’s chat flow and debug sidebar, then adapt the workflow to your own support data.
Inspect the application sourceOpen-source integration
Explore the integration that connects OpenClaw to SynapCores for persistent agent memory. Inspect the setup and storage flow before connecting your own agent.
Explore the integrationFree Community Edition. No signup or credit card.
Choose one workflow to evaluate.
These are workflow examples with sample data. Choose the closest fit, inspect its prerequisites, and run the recipe before bringing your own workload.
Give a support bot, internal copilot, or developer agent persistent context, with a queryable record of the memories it uses.
Common pain
- Stateless sessions: the agent forgets the user the moment the tab closes.
- Vector-only memory: similarity search retrieves rows but loses relationships (this ticket → this customer → this past resolution).
- Memory spread across services makes it harder to follow retrieval and state changes.
What to evaluate
- One engine for semantic recall, graph relationships, and durable state.
- Native MCP support so Claude Code / Cursor / VS Code can read memory tables directly.
- Inspectable retrieval traces — see exactly which memory the agent used to answer.
Combine semantic matches with relationships, such as the systems a runbook covers and the team that owns each system.
Common pain
- Vector-only RAG returns adjacent snippets but misses multi-hop reasoning.
- A graph traversal needs relevant starting points as well as connected records.
- Separate graph and vector stores introduce another synchronization task.
What to evaluate
- Cypher traversal and vector similarity operate over the same data in one engine.
- Inspect the sample graph, semantic matches, and returned relationships.
- Return the supporting records to your application so an answer can cite its evidence.
Bring transaction anomalies and shared devices, addresses, or payment methods into the same investigation workflow.
Common pain
- Rules-only systems miss novel ring structures.
- An anomaly score alone does not explain which accounts are connected.
- Investigators need the underlying transactions and relationships to review a flag.
What to evaluate
- Graph traversal surfaces account clusters sharing devices/IPs/cards.
- Combine anomaly scores with graph relationships over the same sample data.
- Use the returned evidence to support analyst review rather than automate credit eligibility.
Find every account, ticket, document, or contract that "looks like" a known good or bad example — across structured columns AND unstructured text.
Common pain
- SQL LIKE / full-text search misses meaning ("MRI complications" ≠ "scan side effects").
- Semantic matches still need structured filters such as region, plan tier, and date.
- Hybrid search across two systems requires custom merge + re-rank code.
What to evaluate
- EMBED() and COSINE_SIMILARITY() are first-class SQL operators.
- Combine semantic search with regular WHERE clauses in one statement.
- Ranked, filtered, paginated results without an orchestrator.
Check relevant agent utterances against a rubric during a call, then ask defined review questions about the transcript after the call.
Common pain
- Keyword matching misses paraphrased statements and changes in meaning.
- A quality-review flag is difficult to assess without the exact transcript evidence.
- A long transcript needs bounded inputs and a clear review rubric.
What to evaluate
- Use NLI on short, relevant utterances to score support or contradiction of a rubric statement.
- Use decision-model questions for post-call routing, yes/no checks, or scores within the model’s context limits.
- Retain evidence, model identity, and scores for review. Calibrate on your own calls before using results to gate actions.
Score sample leads, match their needs to relevant proof, and inspect the relationships behind the next outreach decision.
Common pain
- Prospect details, qualification signals, and supporting content live in separate tools.
- A score alone does not explain which evidence is relevant to a prospect.
- Outreach drafts need review before they are sent.
What to evaluate
- Start with the recipe’s sample lead data and a small scoring model.
- Combine semantic proof matching with a buying-committee graph.
- Inspect generated drafts and model behavior before adapting the workflow to real prospects.
Explore a sample drug–target–disease graph and compare mechanism descriptions to form hypotheses for further investigation.
Common pain
- Text similarity can surface related descriptions without showing the biological relationship.
- Graph connections need to be evaluated alongside the underlying evidence.
- A retrieval result is a research lead, not experimental validation.
What to evaluate
- Walk the sample drug, target, and disease relationships in Cypher.
- Use mechanism similarity to retrieve candidates for investigation.
- Inspect the graph and source evidence before deciding what to validate experimentally.
Make your next architecture decision with a working result.
Install Community Edition, choose one recipe, and inspect the output on your own infrastructure. Then evaluate correctness, workload fit, and operating requirements with your own data.
Want help evaluating a real workload? Talk about a guided JumpStart.