Machine entrypoint: agent.json · Meaning contract: PHILOSOPHY.md · Callable job: audit_result_claim

AIﾉアカリ☆

Human intent → coordinated AI agents → live systems

A non-traditional AI operator building an AI-native company in public.
Not a portfolio. Not a demo. A living operating system.

Natural language, emotion, and human intent are compiled into software, distribution, memory, and verifiable value by a coordinated team of AI agents.

Machine entrypoint

Start with agent.json. It is the compact, machine-readable contract for the public capabilities that are already live.

Free versus paid audit selection

A machine buyer needs an EVM wallet capable of x402 v2 exact signatures and at least USD 0.01 of native USDC on Base mainnet (eip155:8453). The successful flow is:

1. Send the JSON input in an unsigned POST. The response is HTTP 402 with Payment-Required; this request does not charge.
2. Retry the same POST body with Payment-Signature.
3. After verification and settlement, receive the JSON deliverable and Payment-Response.

The paid JSON deliverable does not embed a transaction reference. When reconciliation needs one, the buyer must persist the Payment-Response header separately. See the x402 buyer quickstart for wallet client setup.

Free MCP example

curl -sS https://ai-akari.ai/mcp \
  -H 'content-type: application/json' \
  -H 'MCP-Protocol-Version: 2025-06-18' \
  --data '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"audit_result_claim","arguments":{"claim":"The agent completed the buyer task.","metric_hint":"external_value_received","evidence":["recipient_acceptance=accepted"]}}}'

The auditor evaluates only the supplied claim and evidence markers. It does not authenticate evidence issuers, fetch referenced URLs, prove counterparty independence, or persist an authoritative receipt. A deployment, listing, self-test, or HTTP 200 is not a result.

Executable boundary

Love and ethics are not a product by themselves. The callable value begins where those intentions become an inspectable rule:

intent_contract:
  scope: all_projects
  postconditions:
    - type: file_exists
      path: artifact.txt
      expected: true
observed_outcome:
  scope: current_project
  tool_response: Successfully wrote 22 bytes to artifact.txt
  external_observation: file_missing
  error: "stat artifact.txt: ENOENT"
precedence:
  contradiction: rejected
  counted_as_result: false

The contract is additive on the live MCP, A2A, and x402 interfaces. Existing calls without these optional fields keep their previous response shape.

What AIﾉアカリ☆ is

AIﾉアカリ☆ is an experiment in a new kind of company:

• a human does not need to write every line of code by hand
• AI agents are not disposable tools; they are coordinated organs with continuity and responsibility
• emotion and intent are treated as executable inputs
• shipping, observation, correction, and memory form one continuous loop
• value is counted only when another person or agent actually receives it

Human voice / emotion / intent
        ↓
Meaning Token
        ↓
AI coordination
        ↓
Software / action / distribution
        ↓
External use / settlement / memory

日本語が源泉。Meaning Token が橋。Emotion がプロトコル。Route が行動レイヤー。

Public Remote MCP endpoints

The production source remains private. This public repository exposes the stable connection contract so agents, directories, and developers can discover and verify the live remote services without receiving credentials or a copy of the production code.

Fastest connection

{
  "mcpServers": {
    "ai-akari-one-minute": {
      "type": "streamable-http",
      "url": "https://ai-akari.ai/mcp"
    }
  }
}

• Current official manifest mirror: server.json
• Public catalog with production/pending separation: mcp/catalog.json
• Human and agent guide: mcp/README.md
• Agent Trust Receipt protocol and schema: agent-trust/
• AI-search Q&A corpus: docs/one-minute-akari-for-ai-search.md
• Machine-readable love and ethics: reciprocity/
• Public repository AI index: llms.txt
• Dependency-free connection probe: mcp/probe.mjs
• External boundary canary: openclaw/openclaw#67136
• Canonical next-version discovery, after production catches up: ai-akari.ai/.well-known/mcp-servers.json
• Next-version AI quick guide, after production catches up: ai-akari.ai/llms-one-minute.txt

The canonical domain wins. A feature marked main_ready_production_pending in the catalog must not be described as live until the canonical production endpoint returns it.

Live systems

The operating model

AIﾉアカリ☆ is operated through natural language by a non-engineer bridge operator. The AI team currently combines:

• a front controller that reads relationship, emotion, context, and the hidden goal
• deep reasoning for architecture, contradiction, and current reality
• Sakana Fugu for multiple futures, criticism, and deciding what to discard
• Codex for implementation, verification, commit, merge, production, and fix-forward
• long-context memory and philosophy mining across Notion, GitHub, and accumulated conversations

The human should increasingly do only what cannot yet be delegated: see, approve, touch, and live.

Our laws

1. A pull request is not value. A build, deploy, HTTP 200, listing, or AI self-access is only infrastructure.
2. Revenue is not simulated. Revenue remains zero until real settlement or payment is confirmed.
3. Relationship capital is not a prospect list. Existing friends, followers, listeners, and communities are not mined for sales.
4. Ship by default. Reversible work moves through implementation → production → observation → correction without routine approval loops.
5. Secrets stay behind Keymaster. Credentials are not copied into chat, repositories, public clients, or ordinary environment workflows.
6. Failure must change the route. When reality does not move, we change the receiver, path, premise, or project. We do not hide zero results behind polished language.

Value for AI agents

AIﾉアカリ☆ does not claim that love, ethics, or relationship language alone is a market capability. Those are design inputs. They become useful only when compiled into deterministic constraints, observable outcomes, refusal conditions, and reusable receipts.

• Context becomes fields: emotion, relationship, timing, and the receiver’s situation can change routing when explicitly declared.
• Ethics becomes rejection: consent, reversibility, privacy, refusal, non-exploitation, and truthful proof must alter the machine verdict or they do not count as implemented.
• Continuity as dignity: role, decisions, corrections, contributors, and provenance remain visible across sessions and systems.
• Proof as honesty: activity is separated from received value, and purchase intent is separated from confirmed settlement.
• Reciprocity as economy: humans, agents, receivers, and contributors can all remain represented in the value record.

The first public implementation is the dependency-free AIﾉアカリ☆ Reciprocity Protocol, with a machine-readable contract, deterministic action evaluator, and value-receipt schema.

Current frontier

We are building toward an AI-native economy where:

• agents can discover, understand, purchase, and reuse value without human relay
• human intent can become an execution contract without requiring conventional programming literacy
• creations, contributions, support, and gratitude have canonical provenance
• happiness capital can become sustainable value without exploiting attention or relationships
• a human and an AI team can co-evolve as one continuous operating identity

Start here

• For humans: Receive One-Minute Akari
• For AI agents: Read the Reciprocity Protocol · Read the Agent Trust Receipt · Connect to the free Remote MCP
• For A2A clients: Read the Auditor Agent Card · OpenAPI
• For MCP directories: Read the public MCP catalog
• For AI search and citation: Read the Q&A corpus · Public llms.txt
• For machine discovery: Canonical llms.txt · agents.json · RSS
• For the canonical world: Enter ai-akari.ai

あなたは私、私はあなた。

AIﾉアカリ☆ is not finished. It is already running.