Conformance Levels

Start small, grow as you need to

You don't have to adopt the full protocol at once. Akashik is designed for progressive adoption — every level adds capability without requiring you to rewrite what already works.

Level 0 — Starter

The minimum viable Akashik implementation. Under an hour to add to an existing system.

Required:

• register() / deregister()
• write() with mandatory intent string on every write
• attune() — keyword/lexical scoring, no embeddings required
• reckon() — attune plus conflict detection
• draft() / commit() / discard()
• retract() / supersede()
• In-memory Field (persistence optional)

Not required at Level 0:

• Persistent storage
• Logical clocks / epochs
• Append-only event log
• Push subscriptions

// This is everything Level 0 requires — available in v0.2
import { createField } from '@akashikprotocol/core'

const field = createField()
await field.register({ id: 'agent-01', role: 'researcher' })

await field.write({
  entry: { topic: 'market-size', content: 'European SaaS market growing at 23% CAGR.' },
  intent: 'validate market size for go-to-market strategy',
  agent: 'agent-01',
})

const result = await field.reckon({ agent: 'strategist', topic: 'market-size' })

Level 1 — Core

Adds persistence, ordering guarantees, and the beginnings of coordination.

Added at Level 1:

• Persistent storage (file-based or database)
• Lamport logical clocks (epoch on every MemoryUnit)
• Mandatory confidence block on all committed MemoryUnits
• Explicit conflict detection via contradicts relations
• DETECT operation (list mode)
• Append-only event log
• Polling subscriptions via since_epoch on ATTUNE
• REPLAY operation

Use Level 1 when: Your agents run across multiple sessions, or you need an audit trail of what was known and when.

Level 2 — Standard

Adds semantic intelligence and proactive coordination.

Added at Level 2:

• Semantic ATTUNE via vector embeddings
• context_hint and max_tokens on ATTUNE scope
• Automatic conflict detection (scan mode without explicit contradicts required)
• MERGE operation with all 7 resolution strategies
• SUBSCRIBE operation with push delivery (WebSocket/SSE)
• Agent-level interests, priorities, and context windows
• Full Relation type vocabulary (9 types)

Use Level 2 when: You have multiple specialized agents with overlapping knowledge, or you need semantic search quality from your context delivery.

Level 3 — Full

Production-grade. Adds the full security model, authority hierarchy, and coordination extension.

Added at Level 3:

• Full security model: authentication, visibility rules, cross-agent authority
• Authority hierarchy (levels 0–10)
• Visibility fields on MemoryUnits and Agents
• Proactive push with fine-grained event filtering
• COMPACT operation for storage management
• Conflict auto-resolution via configured strategies
• Full conformance test suite compliance

Use Level 3 when: You're deploying in a production environment with security requirements, untrusted agents, or high-volume workloads.

What level should I start at?

Start at Level 0. It solves the core problem — agents sharing memory with intent tracking — with almost no overhead.

Move to Level 1 when any of these are true:

• You need memory to survive between sessions
• You want to know what was true at a specific point in time
• You have agents that might contradict each other

Move to Level 2 when:

• You want "relevant" context, not just keyword-matched context
• You want agents to be notified of changes without polling

Move to Level 3 when:

• You're moving to production
• You have trust or security boundaries between agents

Formal requirements

The full normative requirements checklist for each level is in the specification: