Build on Nexefiy’s model of work,
not on its tables.

Describe the API — Version, where the OpenAPI document lives, and the scope catalog. No credential needed.

Who am I — The organization, person, role and scopes this credential acts with.

The OpenAPI 3.1 document — Generated from the same operation registry the server enforces. Feed it to any OpenAPI tool.

List API keys — Your own keys; an admin sees every key in the organization. Never the key itself.

Create an API key — Give `scopes` or a `preset` (`read_only`, `full_access`). You cannot grant a scope your role cannot use. The key acts as you, in this organization, and stops working if you leave it. It is returned once.

Retrieve an API key

Revoke an API key — Immediately and permanently. The key keeps its record (for the audit trail) with `status: revoked`.

List objectives — Newest first. Filter by lifecycle status.

Declare an objective — A new objective starts as a `draft` with its Objective Genome. Activate it with `objectives.update` once a person has reviewed it.

Retrieve an objective

Move an objective through its lifecycle — draft → active → paused / achieved / abandoned. Illegal transitions return `conflict`.

Retrieve an objective's graph — The decomposition of an objective into nodes (sub-goals, tasks, milestones) and the dependencies between them.

List intelligence entities — Agents, humans, systems and services that pursue objectives.

Create an entity with its first genome

Retrieve an entity and its live genome

List an entity's genome versions — The immutable lineage, oldest first. History is never rewritten.

Evolve an entity's genome — Give a full `genome`, or a `patch` merged over the live genome. Creates the next immutable version and makes it live.

List capabilities

Register a capability and its first contract

Retrieve a capability with its live contract

Evolve a capability's contract

Check whether an entity may use a capability — Runs the full authority pipeline (grant, contract, risk, budget, objective, approval) without acting. Nothing is executed. An API key can never supply a human approval.

List executions — Newest first. Filter by status or by the objective served.

Start an execution — A durable run. At most once per `idempotencyKey` (the Idempotency-Key header is used when the body has none). Tasks are then admitted one by one through tenant, state, approval, authority and budget checks.

Retrieve an execution

List an execution's tasks

Add a task to an execution — Set `requiresApproval` to park the task until a person approves it in Nexefiy.

Pause an execution — No new task is admitted until it is resumed.

Resume a paused execution — Admission continues where it stopped.

Cancel an execution — Settles the run and cancels its unfinished tasks.

List an execution's approval gates

Request a human approval — Opens an approval gate (optionally on one task). Only a person decides it, in Nexefiy — never an API key.

Retrieve an execution's trace — Every span — tasks, authority and policy checks, budget, approvals, retries — in order.

Read the event stream — Oldest first, from `cursor` (a previous `nextCursor`) or `fromSequence`. Store the last `nextCursor` to resume exactly where you stopped. Filter with `types` (comma-separated types or families such as `objective.*`).

List event types — Every event type Nexefiy emits — the vocabulary for event filters and webhook subscriptions.

Retrieve an event

List memories — Active memories, newest first. `type=negative` lists negative knowledge — what was tried and failed.

Record a memory

Retrieve a memory

Search memory — Ranked by relevance (pgvector, when `queryEmbedding` is given), recency, confidence, scope and importance — within a privacy ceiling.

List simulation scenarios

Create a scenario — A "what if": a world snapshot, assumptions, variables and the actions under test. Mark one `baseline` (do nothing) to compare against.

Retrieve a scenario with its runs and outcomes

Record a simulation run — Stores the model, inputs, uncertainty and every predicted outcome. Predictions are always marked as simulated, never as fact.

Record what actually happened — Sets the actual value beside a predicted outcome, so the accuracy of every simulation is measured.

List evaluations

Record an evaluation — Expected against actual, with measured metrics. The verdict feeds trust, memory and evolution.

Retrieve an evaluation with its metrics

Performance by subject — Evaluations rolled up per subject (entity, model, capability, strategy …).

Retrieve the Constitution in force — The ratified charter every action is judged against. It changes only by a person ratifying an amendment.

List policies

Retrieve a policy

Ask governance for a decision — Judges a proposed action against the Constitution, every active policy, approval rules and kill switches, and records the decision: `allow`, `require_approval`, `escalate` or `deny`. Nothing is executed. An API key cannot claim a human approval.

List governance decisions

Retrieve a governance decision

List escalations

Raise an escalation to a person

List kill switches

Engage a kill switch — Stops execution in scope immediately (global, an environment, an entity or a capability). Releasing it is a human decision.

List Reality Bridge connections

Retrieve a connection — Credentials are never returned; `credentialConfigured` says whether one is in place.

List the operations the Reality Bridge supports

List external actions

Request a governed external action — Runs the full Reality Bridge pipeline: the connection allow-list, authority grant, Constitution and policies, budget, egress guard, then execution and verification. High-risk actions come back `awaiting_approval` until a person approves them in Nexefiy; forbidden ones come back `denied` with the stage that refused them.

Retrieve an external action

List webhook endpoints

Create a webhook endpoint — Subscribe an https URL to event types (`*`, `objective.created`, or a family such as `run.*`). The response carries the signing secret — the only time it is shown.

Retrieve a webhook endpoint

Change a webhook endpoint — Change its URL, description or event types, or disable and re-enable it. Deliveries for a disabled endpoint wait until it is enabled again.

Delete a webhook endpoint — Stops all deliveries to it, including pending retries.

Rotate an endpoint's signing secret — Issues a new secret. For 24 hours deliveries are signed with both, so receivers can switch without dropping any.

Send a test delivery — POSTs a signed `webhook.ping` to the endpoint now and reports how it answered.

List webhook deliveries — Every attempt to deliver an event, with its status, attempts and the last response.

Retrieve a webhook delivery

Deliver again — Queues a failed, dead or already delivered event for one more attempt, now.