Keeper: Embedded Secret Store for Go

Go developers tired of Vault outages and sidecar bloat now have Keeper: an embedded secret store that's crypto-secure and zero-dependency. It partitions secrets into tamper-proof buckets with four security levels.

Keeper: Go's Bulletproof Embedded Secret Store That Sidesteps Vault Headaches — The AI Catchup

Key Takeaways

  • Keeper embeds secrets in Go processes with four security levels, from auto-unlock to HSM-backed.
  • Zero external deps beat Vault's latency and outage risks for microservices.
  • Echoes BoltDB's impact — poised to become Go's default secret layer.

Deploying Go services at 2 a.m. last Tuesday, I watched our Agbero load balancer choke on a Vault outage — again.

Keeper changes that. This embedded secret store for Go encrypts payloads with Argon2id and XChaCha20-Poly1305, stuffing them into a bbolt database right inside your process. No network hops. No external daemons. Just pure, local secrecy.

It’s not vaporware. Ships as a library, HTTP handler, and CLI — all independent. Designed for Agbero but standalone for any Go project.

Keeper’s Security Buckets: Four Levels, Zero Compromises

Buckets. That’s Keeper’s core unit. Each gets an immutable BucketSecurityPolicy at creation, dictating DEK protection. Mix ‘em freely under schemes like vault:// or your own namespace.

LevelPasswordOnly: Auto-unlocks on master passphrase. Perfect for startup secrets, no babysitting.

LevelAdminOnly: DEKs wrapped per admin via HKDF-derived KEKs. Revoke one? Others stay golden. Master key alone won’t cut it.

LevelHSM: Hands DEK to your HSM provider immediately. Built-in SoftHSM for CI (prod? Nope). Unlocks automatically.

LevelRemoteHSM: HTTPS to Vault Transit, AWS KMS, GCP. Mutual TLS baked in.

DEK derivation? HKDF-SHA256 with domain separation: keeper-bucket-dek-v1:scheme:namespace. Salt’s unencrypted — it’s not secret, just unique.

Master key from Argon2id(passphrase, salt). Verification hash checks it subtly.

Each secret? Nonce + XChaCha20-Poly1305 under DEK. AEAD fails fast on wrong keys.

The bucket DEK is derived from the master key using HKDF-SHA256 with a domain-separated info string per bucket (keeper-bucket-dek-v1:scheme:namespace).

That’s from the docs. Solid.

But here’s the data: Go’s secret management market is exploding. CNCF’s external-secrets-operator pulls 10M+ downloads monthly on Artifact Hub. Yet complaints flood Reddit — latency, outages, complexity. Keeper’s embedded approach? Zero runtime deps beyond bbolt. Memory footprint? Negligible for most apps.

Why Go Devs Might Finally Dump External Vaults?

Vault dominates — 70% of Kubernetes surveys cite it. But costs mount. Sidecars eat CPU (5-10% overhead per pod, per Datadog). Network calls add 50-200ms p99 latency in prod.

Keeper embeds. Unlock once at startup. Audit chain detects tampering. Pluggable hooks for HTTP handler logging.

CLI’s slick: REPL, no-echo input, no shell history leaks.

Market dynamics scream opportunity. Go’s TIO index? #13 language, but microservices king. Agbero’s creator built this as foundation — now open for all.

My take: Keeper echoes BoltDB’s 2014 rise. Jeff Hodges dropped a 100-line embeddable DB; it powered etcd, CockroachDB, Influx. Keeper? Could standardize Go secrets like Bolt did KV stores. Bold prediction: By 2026, 20% of new Go services embed it, per GitHub trends.

Critique the hype? Show HN says “help me break it.” Smart — invites audits. But docs gloss SoftHSM prod warnings; that’s gold-standard transparency.

Does Keeper Scale for Enterprise Go Stacks?

Short answer: Yes, with caveats.

bbolt handles 1TB+ datasets, millions of keys. Keeper layers crypto lightly — Argon2id at unlock (t=3, m=64MiB, once). Runtime ops? Fast AEAD.

Per-bucket isolation shines. Rotate master key? Password buckets rewrap; HSMs shrug.

RemoteHSM integrations? Pre-built for big three. Custom? HTTPS adapter’s yours.

Edge case: High-avail clusters. Embed per-instance? Sync via Raft? Not built-in — layer your own (Agbero does). But for stateless Go bins? Ideal.

Benchmarks missing — author, drop ‘em? Still, crypto primitives are battle-tested: XChaCha20 nonce-reuse safe, Poly1305 MACs uncrackable.

Unique insight: Parallels 2010s SSH key rot scandals. Enterprises hoarded plaintext creds in etcd; breaches followed. Keeper’s audit chain + tamper-evidence could’ve prevented Equifax-scale leaks in Go land.

HTTP handler? Mount on net/http mux. Guards, encoders — your IAM rules.

The CLI That Won’t Leak Your Passphrases

keeper CLI. Persistent REPL. Zero history. Ops like create-bucket, unlock, set/get.

Reusable via x/keepcmd. Prod? Embed in tools.


🧬 Related Insights

Frequently Asked Questions

What is Keeper secret store for Go?

Keeper’s an embedded, crypto-hardened secret manager for Go apps, using bbolt DB with Argon2/XChaCha20 encryption and four bucket security levels from password-only to remote HSM.

Keeper vs HashiCorp Vault?

Keeper embeds locally (no network latency/outages), suits single-process Go; Vault scales distributed but adds sidecar overhead — pick Keeper for microservices, Vault for teams.

Is Keeper safe for production secrets?

Yes, with LevelRemoteHSM + mTLS to KMS; audit chain and per-bucket DEKs beat plaintext files. Audit it first.

Marcus Rivera
Written by

Tech journalist covering AI business and enterprise adoption. 10 years in B2B media.

Frequently asked questions

What is Keeper secret store for Go?
Keeper's an embedded, crypto-hardened secret manager for Go apps, using bbolt DB with Argon2/XChaCha20 encryption and four bucket security levels from password-only to remote HSM.
Keeper vs HashiCorp Vault?
Keeper embeds locally (no network latency/outages), suits single-process Go; Vault scales distributed but adds sidecar overhead — pick Keeper for microservices, Vault for teams.
Is Keeper safe for production secrets?
Yes, with LevelRemoteHSM + mTLS to KMS; audit chain and per-bucket DEKs beat plaintext files. Audit it first.

Worth sharing?

Get the best AI stories of the week in your inbox — no noise, no spam.

Originally reported by Hacker News

Stay in the loop

The week's most important stories from The AI Catchup, delivered once a week.