kernel ~/main

[s] earch

▲ > kernel.start()

Build the runtime everyone inherits.

Kernel gives engineering teams the deploy primitives, observability, and runtime guarantees of a hyperscaler — in twenty minutes flat. Functions, queues, KV, cron, and evals on one inspectable runtime.

● 99.997% uptime

· SOC 2 Type II · 38 regions

~/projects/kernel — zsh 80×24

$ kernel deploy --prod

[10:42:01] INFO Building project kernel@3.12.0... [10:42:03] INFO Resolved 38 regions [10:42:04] INFO Bundling 14 functions, 3 queues, 1 KV [10:42:05] WARN cold-start budget at 92% (12ms)

✓ fn:checkout deployed → ord1, fra1, sin1

✓ queue:invoices deployed → 12 partitions

✓ kv:sessions linked

▲ Deployed in 4.2s

$ _

[OK] 99.997% uptime

[OK] 12ms p50 cold-start

[OK] 38 regions

[INFO] 4.1bn invocations / day

// PRIMITIVES

┌─ Six pieces. │ Six docs pages. └─ Zero abstractions.

functions()

Functions

Sub-12ms cold starts. JS, Python, Go, Rust. Stream tokens out of the box.

→ docs/functions

queues()

Queues

Durable, partitioned, exactly-once. Replay any window. Backfill in seconds.

→ docs/queues

kv()

KV

Globally replicated key-value. Strong reads in-region, eventual everywhere.

→ docs/kv

cron()

Cron

Cron with retries, audit log, and a UI. Per-region or follow-the-sun.

→ docs/cron

logs()

Logs

Structured, searchable, queryable. Tail prod from the CLI in under a second.

→ docs/logs

eval()

Eval

Wire model evals into CI. Block deploys on regression. Score every PR.

→ docs/eval

Trusted by builders.

> █ Atlas

> █ Northwind

> █ Helio

> █ Forge

> █ Continuity

> █ Nimbus

// 142 teams shipping today

// CHANGELOG

Recent ships.

Full log →

2026·05·01

v3.12

Cold-start p50 down to 12ms — 38% faster than v3.11.

2026·04·22

v3.11.2

Patched a queue replay edge case for partitions > 64.

2026·04·14

v3.11

New eval primitive: block deploys on model regression.

2026·04·02

v3.10

KV strong reads across all 38 regions, no opt-in flag.

// CONFIG

Configured by you. Inspected by anyone.

No black boxes. Every deploy is one config file, one diff, and one log stream away from the next engineer who reads it.

~/project/kernel.config.ts TS

01 import { defineConfig } from '@kernel/core'

03 export default defineConfig({

04 runtime: 'edge',

05 regions: ['ord1', 'fra1', 'sin1'],

06 observability: {

07 logs: { level: 'info', sampling: 1.0 },

08 traces: { exporter: 'otel', endpoint: env.OTEL_URL },

09 eval: { onRegression: 'block' },

10 },

11 queues: {

12 invoices: { partitions: 12, retries: 5 },

13 },

14 })

~/project — zsh ▶ run

$ kernel deploy --prod

[10:42:01] INFO building kernel@3.12.0

[10:42:04] INFO bundled 14 fn · 3 queues · 1 kv

✓ fn:checkout → ord1 fra1 sin1

✓ queue:invoices → 12 partitions

✓ kv:sessions → linked

▲ Deployed in 4.2s

// PRICING

Pay for invocations. Not for the runtime.

No per-seat surcharge. No "contact us for pricing." All limits in the open.

HOBBY For weekends and evenings.

$0 /mo

+ 100k invocations / mo + 3 regions + Community Discord + Logs · 7-day retention + Single seat

[ recommended ]

PRO For shipping teams.

$20 /mo

+ 10M invocations / mo + All 38 regions + Priority support · 4h SLA + Logs · 30-day retention + 10 seats

SCALE For production scale.

$80 /mo

+ Unlimited invocations + Dedicated regions + 24/7 on-call · 1h SLA + Logs · 1-year retention + SSO + audit log

[READY]

> ship.kernel()

Twenty minutes from `git init` to a fan-out queue running in 38 regions.

// no credit card · 60-second deploy