Seven databases.

One decision.

cachesessionagent-memoryrate-limiterqueuefeature-flagaudit-log

Each purpose shapes the engine, TTL, eviction, and durability. You declare intent. The platform derives configuration.

Purpose profiles

01cache

02session

03agent-memory

04rate-limiter

05queue

06feature-flag

07audit-log

HOT

COLD

Keys evict as they cool → LRU

Fast and forgettable

Shallow and wide. AkashaKV dual-mode keeps the hot set in cache while the LSM tree preserves writes. Ideal for API response caching, computed values, and CDN origin shields.

engine: AkashaKVttl: 300s defaulteviction: TTL/LRU policydurability: WAL + LSM

Why purpose matters

Generic KV vs. purpose-shaped.

TTL policy

You set it per key. Or forget.

Derived from purpose. Cache: 300s. Session: 24h. Agent-memory: never.

Eviction

Global maxmemory-policy. One size for everything.

LRU for cache. TTL for session. FIFO for queue. None for agent-memory.

Durability

RDB snapshots or AOF. Same for all data.

AkashaKV WAL + LSM persistence for every purpose.

Engine

One engine. Redis.

AkashaKV dual-mode backing engine with purpose-specific keyspaces.

Identity

Connection-level AUTH. Same permissions for all keys.

Per-principal scoping. Agents only access their own keyspace.

Generic KVPurpose-shaped

CLI

One command. Purpose-shaped.

zaps

# Create three databases with different purposes

$ zap create cdn-cache --purpose cache

✓ Created cdn-cache · engine: AkashaKV · ttl: 300s · eviction: TTL/LRU

$ zap create cortex --purpose agent-memory

✓ Created cortex · engine: AkashaKV · ttl: none · eviction: none

$ zap create ops-log --purpose audit-log

✓ Created ops-log · engine: AkashaKV · ttl: 7y · eviction: append-only

# Three databases. Three purposes. Zero configuration.

A cache should not be configured like an audit log.

That is not flexibility. That is negligence.

Next: Agent Identity →See the full platform