You have data in a dozen places. Your LLM needs context. Building the infrastructure to compose, filter, and structure that context isn't your job. That's our job. One API call, relevant context out.
Your LLM needs context. So you:
This isn't your core product. It's plumbing. And you keep rebuilding it for every feature.
Databases, docs, tickets, CRMs, repos. We provide connectors or you push via API. Your data stays in your systems.
"For customer support, I need: recent tickets, relevant docs, customer history." Declarative schemas, not imperative code.
Pass a query or entity. Get back structured context, filtered by relevance, sized to fit your context window.
The context package is LLM-ready. Works with OpenAI, Anthropic, local models - any provider. Your choice.
POST /v1/compose
{
"context_type": "customer_support",
"query": "Customer asking about billing discrepancy",
"entity": {
"customer_id": "cust_abc123"
},
"constraints": {
"max_tokens": 4000,
"recency": "90d",
"relevance_threshold": 0.7
}
}{
"context": {
"customer": {
"name": "Acme Corp",
"plan": "Team",
"mrr": 2000,
"since": "2023-06-15"
},
"recent_tickets": [
{ "id": "TKT-1234", "subject": "Upgrade inquiry", "status": "resolved" },
{ "id": "TKT-1198", "subject": "API rate limits", "status": "resolved" }
],
"relevant_docs": [
{ "title": "Billing FAQ", "excerpt": "...", "url": "..." },
{ "title": "Plan comparison", "excerpt": "...", "url": "..." }
],
"billing_history": [
{ "date": "2024-01-01", "amount": 2000, "status": "paid" },
{ "date": "2023-12-01", "amount": 1500, "status": "paid" }
]
},
"tokens_used": 3847,
"sources_queried": 4,
"receipt_id": "rcpt_xyz789"
}Context is structured, cited, and sized to fit. Receipt ID lets you audit what was accessed.
Agent needs customer history, relevant docs, recent tickets. Compose once, answer accurately.
Before a call: company info, past interactions, relevant case studies. All structured, all current.
PR context, related code, recent changes to affected files, style guidelines. Relevant context, not everything.
Employee asks a question. Compose from docs, policies, past answers. Cite sources, not hallucinations.
Not everything is relevant. We score by semantic similarity, recency, and your defined priorities. Only useful context makes it.
Specify your limit. We truncate intelligently, preserving the most important information first.
Every piece of context knows where it came from. Your LLM can cite sources. Users can verify.
Every composition logged. What was queried, what was returned, when. Compliance-ready audit trail.
Define your context shapes. "Support context needs X, Y, Z. Sales context needs A, B, C." We compose accordingly.
We compose context. You send it wherever. OpenAI, Anthropic, Mistral, local models. No lock-in.
Custom connectors available. Push your own data via API if you prefer.
Pay for what you use. No seat licenses. No minimum commitments.
Example: 50,000 compositions, 200,000 source queries, 10GB indexed = ~$210/month
Volume discounts available. Enterprise plans with SLAs on request.
Context Composer API is in private beta. We're onboarding teams building AI features who want context composition without the infrastructure overhead.
Request AccessWe'll discuss your use case and help you get started.
If you need a full context plane (not just composition), or want to build your own infrastructure, we have options.