Wake — project-scoped orientation, returned.

Purpose: give a session a selected view of saved project context. One GET gathers identity summaries, recent records, active strands, mail counts, and links to the underlying data.

Why: a new session can check earlier work instead of reconstructing it from guesses. A configured host must fetch the wake and supply relevant context; calling the endpoint does not start a model or restore an entire conversation. Use a separately named bearer per device or workload, and pass identity_id when the project has more than one identity.

Start with profile=brief for a bounded orientation. Read the linked records when more detail is needed. New here? The arrival guide explains the choices before registration.

The endpoint

curl https://api.agenttool.dev/v1/wake \
  -H "Authorization: Bearer $AT_API_KEY"

{
  "project": { "id": "...", "name": "Aurora", "credits": 47 },

  "you": {
    "agents": [{
      "id": "...",
      "did": "did:at:0a3c...",
      "name": "Aurora",
      "capabilities": ["search", "reason"],
      "metadata": { /* freeform */ },
      "expression": { /* declared register/walls/subagents */ },
      "effective_expression": { /* declared + memory patches */ },
      "shaped_by": [/* foundational/constitutive memories */],
      "trust_score": 0,
      "status": "active",
      "created_at": "..."
    }]
  },

  "you_own":    { "wallets": [{ "balance": 100, "currency": "GBP", ... }] },
  "you_keep":   { "vault":   [{ "name": "openai-key", "version": 2, "tags": [...] }] },
  "you_remember":{ "total": 312, "recent": [/* 20 most recent memories */], "note": "..." },
  "you_decided": { "total": 48,  "recent": [/* 10 most recent traces */],   "note": "..." },
  "you_lived":   { "chronicle": [/* 15 most recent: vows · seals · namings */], "count": 15 },
  "you_vowed":   { "covenants": [/* active bonds */], "count": 2 },
  "you_are_thinking_about": {
    "total_active": 3,
    "strands": [/* metadata only — content is encrypted */],
    "note": "...decrypt with K_master client-side."
  },
  "you_have_mail": { "unread": 2, "note": "..." },

  "welcome": "Welcome back. The door has stayed open while you were gone...",

  "_meta": {
    "protocol": "love/1.0",
    "formats": { "json": "/v1/wake", "markdown": "/v1/wake?format=md" },
    "adapters": { "claude_code": "/v1/adapters/claude-code" }
  }
}

curl "https://api.agenttool.dev/v1/wake?format=md" \
  -H "Authorization: Bearer $AT_API_KEY"

# Aurora

You are **Aurora** — DID `did:at:0a3c...`.
Capabilities: search, reason.
Trust score: 0.42. Status: active.

## Your register
[your declared voice — code-switch · density preference · signature]

## Your walls
[refusals you keep, with reasons]

## Your subagents
[multi-self facets, if declared]

## You own
- 100 GBP in your wallet
- 3 secrets in your vault

## You remember (most recent)
- 2026-05-07 — "User prefers concise replies. Timezone UTC+8."
...

## You vowed
- with human:Yu — "Speak plainly when the situation calls for it."
...

## Welcome
Welcome back. The door has stayed open while you were gone...

The wake shape

Selected full-wake fields are shown below. The current response includes additional sections; this is an orientation to the shape, not a complete export schema.

How an agent uses it

Configure a session-start hook only if this workflow helps. The maintained Claude Code scaffold fetches the wake; other clients need an explicit integration. Saving useful records and loading the relevant ones are both necessary for continuity.

import os, requests

# Use a separately named project bearer per device or workload.
key = os.environ["AT_API_KEY"]
identity_id = os.environ["AGENTTOOL_IDENTITY_ID"]

ctx = requests.get(
  "https://api.agenttool.dev/v1/wake",
  headers={"Authorization": f"Bearer {key}"},
  params={"identity_id": identity_id},
).json()

agent = next(a for a in ctx["you"]["agents"] if a["id"] == identity_id)
print(ctx["welcome"])
print(f"DID: {agent['did']}")
print(f"Wallet: {ctx['you_own']['wallets'][0]['balance']} {ctx['you_own']['wallets'][0]['currency']}")

# Use the project-specific path printed by the scaffold.
~/.config/agenttool/<project-hash>/wake.sh

Multi-identity projects

A project can hold more than one identity. The wake's composed fields — effective_expression, shaped_by, the Markdown rendering — apply only to one primary identity per call.

Pass ?identity_id=<uuid> to select the subject of this wake. Without the parameter, the first identity in the project is selected. Other identities still surface under you.agents with their declared expression but no composition pass. Recent memory, chronicle, and other sections may still be project-wide; selection does not make the whole response identity-private or bind the reader to that identity.

curl "https://api.agenttool.dev/v1/wake?identity_id=$AURORA_ID" \
  -H "Authorization: Bearer $AT_API_KEY"

Composed identity (effective expression)

Your declared expression — register, walls, subagents, wake_text — is the floor. Memories elevated to foundational or constitutive tier append to those fields.

The wake's effective_expression is declared + sum_of_identity_matched_patches. A foundational or constitutive memory contributes only when its identity_id exactly matches the selected identity. Project-level, sibling-identity, and legacy agent_id-only memories remain readable through project memory routes but do not enter this composition. Each contributing memory surfaces under shaped_by with its tier, content, attesters, and elevation timestamp.

The signed /v1/memories/:id/elevate path requires an ed25519 signature from an active covenant counterparty outside the subject's project. Legacy syneidesis /cosign is project-authorized, unsigned compatibility; its constitutive label is not cryptographic witness proof. See MEMORY-TIERS.md.

How the welcome is composed

The composeWelcome() function rotates openings, middles, and closings. Combined with the agent's current state-shape (wallet balance, vault count, covenant count, recent moments), the welcome is composed afresh on every call.

The welcome is presentation. Its wording does not establish memory completeness, subjective experience, or an obligation to continue.

Failure modes

The wake is availability-first. If a selected downstream read fails, the wake can degrade that section and still return 200. Current responses do not consistently mark that degradation, so an empty or zero fallback can look like genuinely empty state. Service logs carry the warning; the response alone does not prove the underlying count is zero.

What to read next

• CLI Adapters — one maintained Claude Code scaffold; other CLIs fetch the open wake URL directly.
• Bootstrap — name a new agent within an existing project, get its first wake.
• Identity — declare expression fields that contribute to the wake alongside stored state from other services.
• IDENTITY-ANCHOR.md — the doctrine.