ArcFlow Cloud: Edge-Native Graph Mesh

ArcFlow deploys as a mesh of edge nodes — browser, Cloudflare Workers, IoT devices, mobile — that sync via CDC to a GPU-accelerated origin. Not a centralized cloud database.

Where managed cloud graph databases require every query to cross the network to a single data center, ArcFlow runs the same engine at every edge location with sub-millisecond local reads and offline capability. WAL segments ship to object storage for durability; the origin handles GPU-accelerated analytics. Zero cold start (in-process), under 20MB Docker image, no per-query pricing.

The Problem with Centralized Graph Databases#

Traditional managed graph databases follow a single pattern: your graph lives in one data center. Every query crosses the network. Every write has round-trip latency. Offline? You're dead.

This worked when applications lived on servers. It doesn't work when your graph needs to be in:

• The browser — collaborative editing, offline-first apps
• Edge PoPs — sub-millisecond reads at Cloudflare's 300+ locations
• IoT devices — robots, drones, game engines with local graph state
• Mobile apps — offline world models that sync when connected

ArcFlow's Architecture: The Edge Mesh#

ArcFlow takes a fundamentally different approach. Instead of one centralized database, ArcFlow deploys as a mesh of edge nodes that sync via CDC to a GPU-accelerated origin.

              Browser (WASM)
                │
                ▼
     ┌─────────────────────┐
     │   Edge PoP (WASM)    │ ◄── Cloudflare Workers / Deno Deploy
     │   Edge PoP (WASM)    │     Sub-ms reads, local writes
     │   Edge PoP (WASM)    │
     └─────────┬───────────┘
               │ CDC Delta Sync
               ▼
     ┌─────────────────────┐
     │   Object Storage     │ ◄── R2 / S3 / GCS
     │   (WAL Segments)     │     Durable, geo-replicated
     └─────────┬───────────┘
               │
               ▼
     ┌─────────────────────┐
     │   Origin Server      │ ◄── Docker (<20MB), GPU-accelerated
     │   Full ArcFlow       │     PageRank, vector search, spatial
     │   CUDA + Metal       │     Complex analytics, full graph
     └─────────────────────┘

Every node in the mesh runs the same ArcFlow engine. The optimized WASM engine handles reads and local writes at the edge. The origin server handles GPU-accelerated analytics. WAL segments ship to object storage for durability and sync.

Head-to-Head: ArcFlow vs Managed Cloud Graph Databases#

Where managed cloud databases win: managed hosting, multi-region HA, automatic backups, monitoring dashboard. These are ops features, not engine features.

Where ArcFlow wins: everywhere else. Edge, browser, offline, embedded, GPU, temporal, vector, provenance, cost.

How It Works#

1. WAL Segment Shipping#

Every write to ArcFlow generates a WAL (Write-Ahead Log) entry with CRC32 integrity. WAL segments rotate at configurable thresholds (default: 4MB or 10K entries) and ship to S3-compatible object storage.

# arcflow.toml — WAL segment shipping configuration
wal_shipping:
  db_id: "my-graph"
  max_segment_bytes: 4194304   # 4MB
  max_segment_entries: 10000
  backend: s3                  # s3 | gcs | azure | local
  path: "/var/lib/arcflow/wal"

Recovery from object storage: load the latest checkpoint, replay WAL segments after it. CRC32 verified on every segment download.

2. CDC Delta Sync#

Edge nodes sync with the origin using Change Data Capture. Each edge tracks its last-synced sequence number. Deltas are batched and compressed.

Edge Node                          Origin
    │                                │
    │── GET /sync/changes?since=N ──>│
    │                                │
    │<── Delta { ops: [...], seq } ──│
    │                                │
    │── POST /sync/push { ops } ────>│
    │                                │
    │<── ACK { new_seq } ───────────│

Fingerprint verification (db.fingerprint()) ensures consistency after sync.

3. Conflict Resolution#

When multiple edge nodes write concurrently:

• Default: Last-Writer-Wins (LWW) with timestamps
• Pluggable: Custom merge strategy for domain-specific resolution
• Auditable: Conflict log records all resolutions with both versions
• Causal: Vector clocks detect truly concurrent writes vs. ordered writes

4. Deploy in One Command#

# arcflow.cloud.toml
[project]
name = "my-knowledge-graph"
 
[edge]
provider = "cloudflare"
regions = ["us-east", "eu-west", "ap-southeast"]
 
[storage]
provider = "r2"
bucket = "my-graph-wal"
 
[origin]
provider = "fly"
gpu = true
 
# Deploy everything
arcflow deploy

Core Components#

The Positioning#

ArcFlow Cloud is not a cheaper managed graph database. It's a different architecture.

Managed cloud graph databases: "We run a centralized graph database for you."

ArcFlow: "Your graph runs everywhere your users are — browser, edge, mobile, embedded — and syncs to a GPU-accelerated origin for heavy analytics. No vendor lock-in. No per-query pricing. Your data, your infrastructure."

The graph database market is $5.3B and growing 24% annually. Every player offers the same thing: a centralized cloud database. ArcFlow is the only engine that can run at every edge PoP worldwide, sync to a GPU-accelerated origin, and also run offline in the browser.

That's not a feature improvement. It's a category shift.

See Also#

• Sync — WAL-based delta sync, conflict resolution, and db.changesSince()
• Sync Protocol (Deep Dive) — sequence numbers, vector clocks, and merge strategies
• Platform — browser WASM, Docker, Cloudflare Workers, and mobile runtimes
• Agent-Native Database — multi-agent coordination on shared world model state