The portal

The portal at platform.usenexio.com is where your team manages engines, watches runs, reviews results, and issues credentials. The portal shows the same engines, runs and settings your application uses through the API at https://api.usenexio.com. Signing in to the portal does not grant API access: your application always calls the API with an API key (see Authentication and access). What each person sees depends on their role. The tables below name the permission each action needs; Team and roles lists which role holds which permission. The sidebar has two parts: the main areas (Overview, Assistant, Engines, Runs, Conversations, Review, Metrics, and a link to these docs) and Settings (Account, Environments, API keys, Webhooks, Team). An item that needs a permission shows only to people who hold it. Some pages are operated by Nexio; see What Nexio operates for you.

Overview

The first page after sign-in. It summarizes the last seven days.

Part of the page	What it shows	Needs
Summary tiles	Total runs, success rate, 95th percentile latency, failed runs	`runs:read`
Engines	Your organization’s engines	`engines:read`
Recent runs	The three newest runs	`runs:read`
Needs attention	Up to four failed or degraded runs from the last seven days, and, with `review:use`, the number of review items waiting in your Inbox	`runs:read`
Get started	Shortcuts to API keys (with `api_keys:manage`), Webhooks (with `webhooks:manage`), and these docs	none for the docs link

Engines

An engine is a configured model pipeline that turns an input into a structured output (see Engines). The Engines page lists your engines, filters by group, and can show archived engines. Inside an engine, the sidebar groups its pages as Overview, Build, Use, Improve and History.

Page or action	What you do	Needs
Create an engine	Start a new Comparison, Matching or Evaluate engine. Evaluate is the portal’s name for the `entity_analysis` engine type	`engines:manage`
Delete an engine	The engine leaves the list and new runs for it are rejected	`engines:delete`
Contract	Read the request and response contract with examples, and download the engine’s OpenAPI file (`{slug}-openapi.json`)	`engines:read`
Copy for agents	On the Contract page, copy ready-made integration instructions for a coding agent: submit with `POST /api/v1/engines/{slug}/runs`, then poll `GET /api/v1/runs/{run_id}`	`engines:read`
Configuration sections	Edit the engine’s settings. Saving writes the draft. API runs that do not pin `draft` keep executing the latest released version until a new version is published; the Playground and Run pages execute the saved draft, except on engines that require an exact version pin, which refuse them with `engine_version_required`. A few policy settings are read from the saved configuration and apply on save (Settings read from the saved configuration). The sidebar lists the sections only for `engines:manage`	`engines:manage` for every setting, `engines:configure-safe` for the safe subset
Data sources	See whether a connection is bound to the engine, which enrichment sources it calls, and their health	`engines:read` to view; the health data also needs `runs:read`
Playground	Send one real request to the engine and inspect the response before you use it in production. Test keys only	`runs:write`, plus `api_keys:manage` to list the test key it runs with
Run	Run saved example inputs against the engine. Test keys only	`runs:write`, plus `api_keys:manage` to list the test key it runs with
Runs and Metrics	This engine’s runs and its volume and latency	`runs:read`

The safe subset of configuration exists for two engine types. On a Comparison engine it is the scoring dimensions, the weight profiles, the default bucket, built-in scorer parameters and data source. On an Evaluate engine it is the analysis dimensions, overlays and data source labels. Every other setting, including prompts, schemas, redaction settings and enrichment sources, and every change to other engine types, needs engines:manage. A user with only the safe permission who tries to change one gets 403 “Forbidden: safe-config only”.

Publishing a draft from the portal is done by Nexio today. Your integration can publish a new version itself with an organization API key through POST /api/v1/engines/{engine_slug}/versions. Every publish, from either path, passes the same evaluation gate. See Versions and releases and Evaluation.

The Playground and the Run page refuse a live key with “Sandbox only supports test keys”. Runs started there count toward your organization’s monthly run cap, if one is set (see Limits). Operated by Nexio. The Improve and History pages (Evals, Quality, Judge, Proposals, Versions, Lineage) are the learning loop: how Nexio measures an engine against a frozen evaluation set and releases a better configuration. Your team contributes to it with annotations and outcomes; see Outcomes and annotations and Evaluation.

Runs

A run is one execution of an engine on one input (see Runs). The Runs page lists runs for the environment chosen in the environment picker at the bottom of the sidebar, with the columns Run, Engine, Environment, Status, Duration and Created. It pages with a cursor. Opening it needs runs:read. When a run has output, the run detail page has an Export output button that downloads the output your application receives. The page has these tabs:

Tab	What it shows	Who sees it
Result (the label depends on the run’s output)	The run’s output, rendered for reading	anyone who can open the run
Audit	The evidence behind the output	`runs:read`
Request / Response	The request your application sent and the response it received	anyone who can open the run
Review	Annotate the run and record your review decision	`review:use`
Trace	The step-by-step execution trace, up to 1,000 events	`runs:trace`

runs:read lets a user open any run in the organization. A reviewer without runs:read can open only the runs assigned to them through an open review item. Personal data in test runs is masked unless the user holds pii:reveal:test; personal data in live runs stays masked in the portal for customer roles. Operated by Nexio. Nexio sends runs to the review queue, one at a time or in bulk.

Conversations

A conversation instance is a configured assistant your application talks to over the API (see Conversations). The Conversations page lists your instances and lets you monitor every conversation in the organization. Transcripts open read-only.

Page or action	What you do	Needs
View instances, conversations and versions	Browse instances, read transcripts, see released versions	`engines:read`
Create, configure, evaluate and publish an instance	Author the instance, write evaluation scenarios, release a version	`engines:manage`
Chat	Talk to the instance from the portal	`runs:write`
Metrics	Volume and quality for the instance	`runs:read`
Feedback	Read the annotations people left on turns	`engines:read`

Operated by Nexio. Nexio triages conversation feedback. The sidebar’s Assistant item opens a chat assistant at /assistant. Starting it needs engines:read, and sending a message needs runs:write.

Review

Review is where people check results and record what they found (see Review). It needs review:use.

Tab	What it holds
Inbox	Items assigned to you and waiting for review
Snoozed	Items you set aside until later
Done	Items you finished
All	The items of the other three tabs together

On each item you annotate the run, mark it done with a disposition, snooze it, or reopen it. Operated by Nexio. Putting runs into review, assigning items to reviewers, bulk actions and deleting items need review:moderate. Extraction review, where a person verifies facts read from your documents, is also done by Nexio in the connection workspace (see Review).

Metrics

Metrics shows run volume and quality over a time range. It needs runs:read.

Panel	What it shows
Summary row	Totals, success rate and latency for the range
Status breakdown	Runs by status
Daily volume	Runs per day; select a day to open the Runs page filtered to it
Latency	Latency over the range
Slowest runs and recent failures	The runs to look at first

Pick a preset from today to 90 days, or a custom range. The portal remembers the range between visits.

Settings

Portal path	What you do	Needs
Settings, then Account	Edit your first and last name. Your email comes from your sign-in and cannot be changed here	none
Settings, then Environments	Create a sandbox environment, rename one, or delete a sandbox that has no keys, webhooks or runs. The live environment cannot be deleted. Up to five non-production environments; email support@usenexio.com for more	`environments:manage` to open the page, `api_keys:manage` to change environments
Settings, then API keys	Create an organization API key for one environment (prefixed `nx_<environment>_`), rename it, or revoke it. The key is shown once. Revocation cannot be undone	`api_keys:manage`
Settings, then Webhooks	Add an endpoint (environment, URL, events, full or thin payload), watch delivery health, resend a failed delivery, send a test delivery, rotate the signing secret, or delete the endpoint	`webhooks:manage`
Settings, then Team	Invite people, change roles, remove members, see recent membership activity	`members:manage` (changing a role also needs `roles:manage`)

The API keys page issues organization API keys only. Capability-scoped keys (nxsk_v1_...) are issued by Nexio on request; see Authentication and access. For environments, see Environments. For webhooks, see Webhooks.

What Nexio operates for you

Some pages are operated by Nexio. You supply credentials and scope; Nexio configures, validates and monitors.

Capability	What Nexio does	Read more
Connections	Add and configure connections to your document libraries, databases, warehouses, APIs and MCP servers; index and monitor them	Connections
Extraction review	Verify facts extracted from your documents, one document at a time	Review
Run review queue	Put runs into review and assign them to your reviewers	Review
Learning loop	Measure engines against evaluation sets, draft and approve configuration changes, publish from the portal	Evaluation
Data sources catalog	Maintain the list of enrichment sources and the fields each returns	Connections
Usage and billing	Review metered usage and set a monthly run cap on request	Limits

Team and roles

Which role to give each person on your team.

Authentication and access

Issue a key and make your first authenticated call.

Start here

Engines

Conversations

Connections and data

Records

Events and webhooks

Platform

Reference

Overview

Engines

Runs

Conversations

Review

Metrics

Settings

What Nexio operates for you

Team and roles

Authentication and access

​Overview

​Engines

​Runs

​Conversations

​Review

​Metrics

​Settings

​What Nexio operates for you

Team and roles

Authentication and access

Overview

Engines

Runs

Conversations

Review

Metrics

Settings

What Nexio operates for you