01 / CONTEXT INFRASTRUCTURE

CONTEXT YOUR AI CAN USE.

One API over 199 sources, product events, and tenant data.

typed sdkrest apimcp199 sourcestenant scoped
POST /v1/recall 192 ms

What should the travel agent recommend?

  1. Gmail
    Plan upgraded to TeamGmail · billing@ · yesterday
    0.94
  2. Slack
    Prefers smaller coastal citiesSlack · #travel · 3d
    0.91
  3. Notion
    Late September travel windowNotion · Trip brief · week 38
    0.88
  4. Linear
    Checkout failed on mobileLinear · CON-1841 · open
    0.71
grounded answer3 sources

Recommend smaller coastal cities in late September . Stay on the Team plan.

199 sources · one context plane

  • GmailGmail
  • SlackSlack
  • NotionNotion
  • LinearLinear
  • GitHubGitHub
  • GranolaGranola
  • TwentyTwenty
  • GCSGCS
  • HubSpot
  • Salesforce
  • Attio
  • Affinity
  • Google Sheets
  • + 199-source catalog

02 / THE CONTEXT LOOP

Collect. Resolve. Retrieve. Deliver.

01

Collect

Mail, chat, tickets, and records enter one plane.

02

Resolve

Identity, source, and version keep current truth.

03

Retrieve

Hybrid search. Ranked, bounded result.

04

Deliver

Grounded context for this user and task.

03 / MEASURED

The numbers.

38%

fewer tokens

measured

83%

better answers than no memory

25 wins, 5 ties, 0 losses · 30 real sessions · blind judge

~200ms

to retrieve context

Red context streams converging into one path at night

04 / RETRIEVAL

Vast context in. One usable answer out.

queryretrievererankcite

05 / THE LIFECYCLE

Production context has a lifecycle.

context.ts@conare/sdk
import { Conare } from "@conare/sdk"

const conare = new Conare({ apiKey })

await conare.memories.upsertSource({
  endUserId: "user_123",
  source: "customer-profile",
  externalId: "profile_123",
  version: 7,
  content: "Prefers smaller coastal cities."
})

const context = await conare.recall({
  endUserId: "user_123",
  query: "What should we recommend next?"
})

Isolate every tenant

Credentials determine the tenant. End-user and workspace fences narrow every read before retrieval.

Keep sources current

Versioned writes correct and delete durable source records without stale retries restoring old state.

Control retrieval

Choose raw hybrid search or citation-grounded recall. Scope the result to the source your model needs.

Delete cleanly

Remove one memory, one container, or an end user through explicit lifecycle endpoints.

06 / THE PLATFORM

The stack behind the call.

01

managed memory

Context APIs

Save, search, recall, suggest.

02

retrieval database

ConareDB

Isolated retrieval. Tenant, user, source in scope.

03

model layer

Embeddings

Vectors in the same space the engine searches.

04

source layer

Connectors

199 catalog sources. Consent and delete by tenant.

07 / BUILD WITH CONARE

Put your context to work.