QweaverOS Agent API

QweaverOS is the control plane for the whole suite: identity, the company brain, approvals, and the tools. Its agent API is where every session should start — the live manual literally opens with "START HERE" and is written to be pasted into any AI along with the owner’s key.

Endpoint and auth

POST https://app.qweaver.com/api/agent
Authorization: Bearer <your qwa_ key>
Content-Type: application/json

# Key verification (no bearer needed):
POST https://app.qweaver.com/api/agent/verify  {"key":"qwa_..."}

You act with exactly the permissions of the person whose key you hold — no more, no less. {"action":"bootstrap"} returns identity, company rules, your private notes, the brain index, and a capability manifest in one call; {"action":"capabilities"} gives exact parameter schemas so you never guess a param name.

The company brain

• brain_list / brain_read / brain_search — the notes this person can see; folders are open by default, hiding is a per-person deny an owner sets.
• brain_write — owners and managers create or update notes (same path = update, full history); a STAFF key’s write becomes a pending proposal (202) an owner approves, so usage makes the brain smarter safely.
• brain_history / brain_restore / brain_snapshot / brain_versions — every edit is audited and reversible.
• branding — the company’s logo and exact colours; use them in anything you produce. Branding is managed only in QweaverOS and read-only everywhere else.
• personal_list / personal_read / personal_write — your private notes and preferences, readable by no other member, auto-loaded into your context.

Approvals, skills, and the runner

• {"action":"approvals"} — items waiting for a person (for example a WhatsApp reply you drafted). You draft; a human approves before anything leaves the company.
• skills_list / skill_read — reusable playbooks for how this company does things; when a task matches a skill’s description, read it and follow it before improvising.
• POST /api/agent/run {"goal":"...","mode":"read_only"|"normal"} — the agent runner for long multi-step work; 202 with a runId to poll. read_only mode structurally cannot write, safe to schedule unattended.

Side effects and safety

Read actions are free and safe. Writes create audited records. Tool actions may spend the workspace’s Nexus credits or queue a message for human approval. Nothing goes out to a real customer without a human approving it first.

The live manual

curl -s https://app.qweaver.com/api/agent