SolarflareDB
Compare Open demo
Technical-preview product concept

The context database for agents that need to remember.

Store facts, episodes, relationships, and provenance in one temporal context layer. Retrieve by meaning, structure, text, or time—without stitching together a graph database, vector index, and event log.

Run the five-minute demo Read the architecture
Free Spark plan Hard spend caps Proposed TypeScript, Python, and HTTP APIs
vector_indexed_through
us-east:1842
as_of
2026-08-19T18:42Z
Context query
authoritative
What product name did the customer select, and why?
Fact · current0.94

The customer selected SolarflareDB as the canonical brand and owns both related domains.

recorded 18:46valid nowsource msg_1842
semantic
lexical
graph
time
Bitemporal contextRecorded time and valid time
Hybrid retrievalMeaning, text, graph, and time
Deterministic replayInspectable bookmarks and history
Transparent usageReads, mutations, search, storage
One context plane

Memory is more than a similarity search.

Agents need to know what changed, which source supports a claim, how entities relate, and whether a result is current. SolarflareDB makes those questions part of the data model.

Know what is true now

Represent recorded time, valid time, supersession, contradiction, and retraction without deleting the history that explains the change.

Explore temporal context

Retrieve through every signal

Fuse semantic ANN, lexical relevance, typed graph patterns, exact identifiers, temporal filters, and policy into one explainable result.

See the query model

Share context without silent races

Use typed immutable structure, explicit merge policies, regional bookmarks, and replayable operations for multi-agent shared memory.

Review architecture
Replace the stitched stack

One model for the full context lifecycle.

Keep the event history, graph structure, retrieval indexes, temporal interpretation, and provenance aligned around one authoritative operation stream.

Chat and event historyanother store
Vector similarityanother index
Entity graphanother database
SolarflareDB contextone contract
Five-minute activation

From API key to explainable memory.

The onboarding path ends only after a developer writes context, retrieves it, inspects its source, and sets a budget boundary.

01

Create a project and scoped key

The key is shown once and belongs to one project.

02

Commit one episode

Atomic, idempotent, source-attributed, and bookmarked.

03

Query and inspect provenance

See retrieval signals, temporal status, and index watermarks.

import { Solarflare } from "@solarflaredb/sdk";

const db = new Solarflare({ apiKey: process.env.SOLARFLAREDB_KEY });

await db.memory.remember({
  actor: "agent:planner",
  episode: "The customer selected SolarflareDB as the canonical brand.",
  validFrom: new Date(),
  source: { type: "conversation", id: "msg_1842" }
});
from solarflaredb import Solarflare

client = Solarflare(api_key=os.environ["SOLARFLAREDB_KEY"])

client.memory.remember(
    actor="agent:planner",
    episode="The customer selected SolarflareDB as the canonical brand.",
    valid_from=datetime.now(timezone.utc),
    source={"type": "conversation", "id": "msg_1842"},
)
curl -X POST https://api.solarflaredb.com/v2/graphs/memory/mutate \
  -H "Authorization: Bearer $SOLARFLAREDB_TOKEN" \
  -H "Idempotency-Key: msg_1842" \
  -H "Content-Type: application/json" \
  -d '{ "operations": [ { "operationKind": "append-event", "body": { "key": "episode:msg_1842", "event": { "text": "The customer selected SolarflareDB as the canonical brand." } } } ] }'
Usage without ambiguity

Start free. Pay for actions you can explain.

No proprietary memory credits. Spark and Hobby are hard-capped; production plans use visible overages behind required spend ceilings.

Spark
$0/ month

Learn, evaluate, and publish a small demo.

  • 1M reads
  • 25k mutations
  • 10k hybrid searches
  • 250 MB hot storage
  • Hard cap · no card
Open demo
Hobby
$5/ month

Personal agents, portfolios, bots, and student teams.

  • 2M reads
  • 50k mutations
  • 25k hybrid searches
  • 1 GB hot storage
  • Hard cap by default
Open demo
RecommendedLaunch
$29/ month

One production application or a small product team.

  • 10M reads
  • 200k mutations
  • 75k hybrid searches
  • 3 GB hot storage
  • Spend-controlled overage
Open demo
Scale
$199/ month

Growing agent SaaS and multi-agent workloads.

  • 75M reads
  • 1.5M mutations
  • 750k hybrid searches
  • 25 GB hot storage
  • Priority support target
Open demo
Estimate a workload
Policy gatewayIdentity, tenant, region, quota, request limits
→
Transactional graph shardsAuthoritative state, replay, HLC, idempotency, outboxes
→
Derived context indexesR2 archive, lexical, vector, projections, mirrors

The private-beta core now implements predicate-level blue/red routing, atomic Durable Object commits, vector-frontier bookmarks, strict replay, R2 origin logs, and reactive regional materialization. The R2 LSM, complete gluing fanout, ANN, shapes, analytics, and erasure workflows remain measured release phases.

Make context inspectable

Build the agent that remembers what changed.

Walk through the browser-only onboarding demo, then inspect the deployable Cloudflare private-beta code and the evidence-labelled architecture.

Open the product demo Read the quickstart