01Documentation

Everything, in the order you need it.

The documentation set is being written alongside the product. Section outlines are below so you can see the shape of it and tell us which part you need first.

Designed
Quick start · designed interface

Three steps: register the agent, guard the consequential call, honour the decision. Everything else — traces, cost attribution, the audit record — is produced from that.

fromfromimport psifi import Latticelattice = Lattice(api_key=osos.environ["PSIFI_API_KEY"])# Register the agent once, at startup.agent = lattice.agent(    id="invoice-resolution",    environment="production",    owner="finance-automation",)# Wrap the consequential call. Lattice evaluates policy# before the request reaches the payment system.withwith agent.run(trace="tr_8f21c4") asas run:    decision = run.guard(        tool="payments.refund.create",        payload={"order": order_id, "amount": 82_400},    )    ifif decision.requires_approval:        run.wait_for_approval(decision.approval_id)    ifif decision.allowed:        payments.refunds.create(order=order_id, amount=82_400)
DesignedDesigned integration model · the interface is illustrative and subject to change
01

Getting started

4 pages
OverviewWhat the control plane does and where it sits.Core conceptsAgents, runs, traces, policies, decisions, audit.Connect an agentSDK, gateway, MCP or OTLP.First policyWrite a rule in observe-only mode.
02

Architecture

4 pages
Control planeRegistry, policy engine, approvals, audit ledger.Runtime pathWhere an action is intercepted.Event pipelineIngest, queue, store, index.DeploymentReference topology and private deployment.
03

Agents

4 pages
Registering agentsIdentity, owner, environment, risk.Tool permissionsScopes and allow / approve / deny.EnvironmentsDevelopment, staging, production.Inventory APIList and update agent records.
04

Runtime

4 pages
Traces and spansThe shape of a recorded run.AttributesWhat is captured and what is redacted.Live tailWatching runtime events.EvaluationsScoring agent output over time.
05

Policies

4 pages
Policy modelConditions, actions, scope, environment.Actionsallow · deny · require_human_approval · redact.Testing policiesEvaluate against a sample payload.VersioningHow a change is recorded and rolled out.
06

API

4 pages
AuthenticationScoped API keys per environment.ActionsEvaluate and record an action.ApprovalsRequest, list and decide.AuditQuery and export the ledger.
07

SDK

4 pages
PythonInstall, configure, instrument.TypeScriptNode and edge runtimes.Guarding a callThe one function that matters.Error handlingFail-open and fail-closed behaviour.
08

Integrations

4 pages
OpenTelemetrySend existing spans over OTLP.MCPGovern tools exposed over MCP.WebhooksApproval and policy events.GatewaysRouting model traffic through Relay.
09

Security

4 pages
Data handlingRedaction, retention, storage.IdentitySSO, RBAC, API keys.Tenant isolationWorkspace boundaries.AuditWhat is recorded and for how long.
10

Examples

4 pages
Governed refundThe scenario on the home page, end to end.Vendor creation approvalApproval on a write to an ERP.Redacting model contextA data rule applied before inference.Observe-only rolloutIntroducing policy without blocking.
Not published yet

These outlines describe pages that are being written, so they are not links. If a specific one would unblock an evaluation, ask and we will send the draft.