@almyty/models 1.5.0 → 1.5.2

package/README.md

@@ -11,20 +11,25 @@ npx @almyty/models list
 
 ## What makes a model usable
 
-Support in almyty is registry data, never a code list. A model is usable when
-its **card** exists in your organization's catalog and:
+Support in almyty is registry data, never a code list. You connect a provider
+once and every model it lists shows up. A model is usable when its **card**
+exists in your organization's catalog and:
 
 1. something can call it — a stored LLM provider row, or an endpoint URL from
-   a deployment,
+   a hosted model,
 2. its status is `active`, and
-3. one **validation run** has passed: a real, short call, recorded.
+3. it is checked: its provider's key check has passed (one real call with the
+   key covers every model that provider lists), or, for an endpoint with no
+   provider row, a real short call through that model has.
 
-`list` and `get` report which of those is missing, so a card that will not be
-picked says so instead of looking like any other row:
+A model the provider stops listing goes `inactive`; one the vendor answers
+"model not found" for is marked unavailable. `list` and `get` report which of
+those applies, so a card that will not be picked says so instead of looking
+like any other row:
 
 ```
 Llama 3 8B  [llama-3-8b]  private_cloud/eu-central  $0.1/$0.2 per M (feed:litellm)
-    9c2f…  not selectable: no passed validation run (pending) — run: almyty models validate 9c2f…
+    9c2f…  not selectable: waiting for its provider's key check (check the provider on the Models page)
 ```
 
 ## Commands
@@ -36,17 +41,16 @@ Every read command takes `--json` and writes undecorated JSON to stdout.
 | Command | What it does |
 |---|---|
 | `list [--selectable] [--status s] [--tier t] [--provider id]` | Model cards, each line saying selectable or why not |
-| `get <id>` | One card in full: what can call it, capabilities, pricing, the last validation run, measured latency |
-| `register --name <n> --provider <providerId> --model <vendorModelId> [--tier t] [--region r] [--context n]` | Register a card against a stored LLM provider |
-| `register-endpoint --name <n> --url <baseUrl> --model <vendorModelId> [--api-key-stdin] [--tier t] [--region r] [--context n]` | Register any OpenAI-compatible server you run |
+| `get <id>` | One card in full: what can call it, capabilities, pricing, the last check, measured latency |
+| `register --name <n> --provider <providerId> --model <vendorModelId> [--tier t] [--region r] [--context n]` | Register a card against a stored LLM provider. A server you run (vLLM, TGI, llama.cpp) is added as a `custom` LLM provider first, then registered against like any other |
 | `set <id> [--name n] [--tier t] [--region r] [--context n] [--status s] [--price-in n --price-out n] [--clear-price]` | Change a card; a price pair is an override that wins over the automatic feed |
-| `sync [providerId]` | Import what a provider lists, as unvalidated cards. With no id, every active provider |
-| `validate <id>` | One real short call. Passing is what makes a card selectable |
+| `sync [providerId]` | Import what a provider lists; selectable at once when the provider's key check has passed. With no id, every active provider |
+| `validate <id>` | One real short call through one model. A provider's key check already covers its models; this is for an endpoint with no provider row |
 | `delete <id>` | Remove a card |
 
-Cards mostly arrive on their own: creating an LLM provider, changing it, and
-every passing health check import what that provider currently lists. `sync`
-is the same import by hand.
+Cards arrive on their own: connecting a provider, changing it, every passing
+key check and a sweep every six hours import what that provider currently
+lists. `sync` is the same import by hand.
 
 ### Routing
 
@@ -68,7 +72,7 @@ npx @almyty/models route --objective cheapest --tier private_cloud \
 Rejected 3:
   1d0e…  no callable provider
   7bb2…  privacy tier public exceeds the ceiling private_cloud
-  e551…  no passed validation run
+  e551…  not usable yet
 ```
 
 | Flag | Meaning |
@@ -85,7 +89,7 @@ Rejected 3:
 
 It exits 5 when no card satisfies the policy, so a check can be a check.
 
-### Versions, adapters, deployments
+### Versions, adapters, hosting
 
 Registering a version is optional: do it when you want lineage, a manifest
 digest and evaluation history attached to your own artifact. Skip it to just
@@ -96,21 +100,24 @@ run a model that already lives somewhere.
 | `versions` | Registered model versions |
 | `register-version --name <n> --uri <pinned uri> [--base b] [--quantizations q1,q2]` | `hf://org/repo@sha`, `s3://bucket/key@etag`, `gs://bucket/key@gen`, `file:///path@sha` |
 | `adapters` | Every adapter: what it can run (`modelSchemes`), its capabilities, which config fields are secret |
-| `deploy <model> --adapter <key> [...]` | Run a model on a provider's managed product |
-| `deployments` | Desired vs actual, state, spend |
-| `deployment <id>` | One deployment in full, including its endpoint and rate |
+| `host <model> --adapter <key> [...]` | Run a model on a provider's managed product |
+| `hosted` | Hosted models: desired vs actual, state, spend |
+| `hosted <id>` | One hosted model in full, including its endpoint and rate |
 | `scale <id> <replicas>` | Set desired replicas; `0` scales to zero |
 | `teardown <id>` | Tear the endpoint down; weights stay in the registry |
 
+`deploy`, `deployments` and `deployment <id>` also work, as other names for
+`host`, `hosted` and `hosted <id>`.
+
 Naming the model is configuration, so it is the positional argument:
 
 ```sh
-npx @almyty/models deploy hf://Qwen/Qwen3-0.6B@main --adapter huggingface-endpoints
-npx @almyty/models deploy fireworks://accounts/acme/models/qwen3-tuned --adapter fireworks
-npx @almyty/models deploy --model-version <id> --adapter modal --desired '{"replicas":1}'
+npx @almyty/models host hf://Qwen/Qwen3-0.6B@main --adapter huggingface-endpoints
+npx @almyty/models host fireworks://accounts/acme/models/qwen3-tuned --adapter fireworks
+npx @almyty/models host --model-version <id> --adapter modal --desired '{"replicas":1}'
 ```
 
-An **artifact** reference points at bytes and is pinned, so the deployment is
+An **artifact** reference points at bytes and is pinned, so the hosted model is
 reproducible. A **provider reference** (`bedrock://`, `vertex://`,
 `fireworks://`, …) names a model that already exists on a platform, which
 versions it itself. The two do not mix freely: `adapters` lists what each
@@ -123,35 +130,20 @@ provider can really read, and a mismatch is refused at submit with
 history keeps them, and most CI runners echo them. So this tool does not take
 a secret as a flag value.
 
-**An endpoint key.** Prompted without echo, or read from stdin:
-
-```sh
-npx @almyty/models register-endpoint --name vllm-box --url https://vllm.internal/v1 --model llama-3-8b
-# API key for the endpoint (empty for none): ······
-
-pass show vllm/key | npx @almyty/models register-endpoint \
-  --name vllm-box --url https://vllm.internal/v1 --model llama-3-8b --api-key-stdin
-
-# an endpoint with no key at all
-npx @almyty/models register-endpoint --name open-box --url https://box/v1 --model m --api-key ""
-```
-
-`--api-key <value>` is refused, and says this.
-
 **Adapter configuration.** Best is not to paste one at all: connect the
 provider account once and name the connection.
 
 ```sh
-npx @almyty/connections connect huggingface
-npx @almyty/models deploy hf://Qwen/Qwen3-0.6B@main \
+npx @almyty/credentials add huggingface
+npx @almyty/models host hf://Qwen/Qwen3-0.6B@main \
   --adapter huggingface-endpoints --credential <connectionId>
 ```
 
 Otherwise pass the object from a file or stdin:
 
 ```sh
-npx @almyty/models deploy hf://Qwen/Qwen3-0.6B@main --adapter huggingface-endpoints --config-file hf.json
-cat hf.json | npx @almyty/models deploy hf://Qwen/Qwen3-0.6B@main --adapter huggingface-endpoints --config-stdin
+npx @almyty/models host hf://Qwen/Qwen3-0.6B@main --adapter huggingface-endpoints --config-file hf.json
+cat hf.json | npx @almyty/models host hf://Qwen/Qwen3-0.6B@main --adapter huggingface-endpoints --config-stdin
 ```
 
 `--config` still works for the fields an adapter does **not** mark secret, and
@@ -183,7 +175,7 @@ says where a number came from: `feed:litellm`, `feed:openrouter`, `native`,
 | 2 | usage error (bad flags, missing argument, unknown command) |
 | 3 | not authenticated — run `npx @almyty/auth login` |
 | 4 | not found |
-| 5 | the operation ran and failed (a validation run that did not pass, a policy that resolves to nothing) |
+| 5 | the operation ran and failed (a check that did not pass, a policy that resolves to nothing) |
 
 The same table in every `@almyty/*` CLI.
 

package/dist/index.d.ts

@@ -5,6 +5,14 @@ export interface ParsedArgs {
     flags: Record<string, string | boolean>;
 }
 export declare function parseArgs(argv: string[]): ParsedArgs;
+/**
+ * Other names the hosting commands answer to, matching the API's
+ * /model-deployments routes. They work but stay out of --help, which speaks
+ * of hosting. `deployment <id>` is `hosted <id>`.
+ */
+export declare const COMMAND_ALIASES: Readonly<Record<string, string>>;
+export declare function resolveCommand(command: string | undefined): string | undefined;
+export declare function helpText(): string;
 /**
  * A number, or a message. `--context abc` used to become NaN, which
  * JSON.stringify turns into null, so the API saw a field it could not
@@ -26,7 +34,6 @@ export declare function secretFields(schema: any): string[];
 export declare function assertNoArgvSecrets(schema: any, config: Record<string, unknown>, flag: string, alternatives: string[]): void;
 /** Request bodies are built from flags here so they can be checked without a network. */
 export declare function registerBody(flags: ParsedArgs['flags']): Record<string, unknown>;
-export declare function registerEndpointBody(flags: ParsedArgs['flags'], apiKey?: string): Record<string, unknown>;
 /** A card update. Refuses an empty one rather than sending a PATCH that does nothing. */
 export declare function setBody(flags: ParsedArgs['flags']): Record<string, unknown>;
 /**
@@ -36,7 +43,7 @@ export declare function setBody(flags: ParsedArgs['flags']): Record<string, unkn
 export declare function routePolicy(flags: ParsedArgs['flags']): Record<string, unknown>;
 /**
  * Naming the model is configuration, so the model reference is the
- * positional argument: `deploy hf://org/repo@sha --adapter huggingface-endpoints`.
+ * positional argument: `host hf://org/repo@sha --adapter huggingface-endpoints`.
  * `--model-version` is the other way in, for people who registered an
  * artifact to get lineage and evaluation history with it.
  */
@@ -69,3 +76,10 @@ export declare function formatSync(data: any): string;
  * type JSON and press ctrl-D, which looks exactly like a hang. Say so.
  */
 export declare function assertStdinIsPiped(flag: string): void;
+/**
+ * True when `argv1` is this module, directly or through the npm bin
+ * symlink. The old check matched the path against a regex and bailed out
+ * when VITEST was set, so a smoke test that spawned the built CLI from
+ * vitest inherited VITEST and got no output at all.
+ */
+export declare function isEntrypoint(argv1: string | undefined, moduleUrl: string): boolean;

package/dist/index.js

@@ -3,21 +3,21 @@
  * @almyty/models: the model catalog from the terminal.
  *
  * Support in almyty is registry data, never a code list: a model is usable
- * when its card exists, has something that can call it, is active, and has
- * one passed validation run. `list` and `get` say which of those is missing,
+ * when its card exists, has something that can call it, is active, and is
+ * checked (its provider's key check passed). `list` and `get` say which is missing,
  * and `route` answers the question a list cannot — what a routing policy
  * would pick right now, and why it rejected the rest. See docs/models.md.
  *
  * Adapter configuration and endpoint keys are secrets, so they are never
  * taken from argv: argv is readable through `ps` and lands in shell history.
  */
-import { createInterface } from 'readline';
-import { readFileSync } from 'fs';
+import { readFileSync, realpathSync } from 'fs';
+import { fileURLToPath } from 'url';
 import { AlmytyClient, resolveCredentialsOrExit } from '@almyty/client';
 import { EXIT, EXIT_CODE_HELP, UsageError, describeError, exitCodeFor } from './exit-codes.js';
 import { VERSION } from './version.js';
 /** Flags that never take a value, so they never swallow the next argument. */
-const BOOLEAN_FLAGS = new Set(['json', 'selectable', 'config-stdin', 'api-key-stdin', 'clear-price']);
+const BOOLEAN_FLAGS = new Set(['json', 'selectable', 'config-stdin', 'clear-price']);
 export function parseArgs(argv) {
     const result = { positional: [], flags: {} };
     for (let i = 0; i < argv.length; i++) {
@@ -69,12 +69,29 @@ export function parseArgs(argv) {
     }
     return result;
 }
+/**
+ * Other names the hosting commands answer to, matching the API's
+ * /model-deployments routes. They work but stay out of --help, which speaks
+ * of hosting. `deployment <id>` is `hosted <id>`.
+ */
+export const COMMAND_ALIASES = {
+    deploy: 'host',
+    deployments: 'hosted',
+    deployment: 'hosted',
+};
+export function resolveCommand(command) {
+    return command && Object.prototype.hasOwnProperty.call(COMMAND_ALIASES, command) ? COMMAND_ALIASES[command] : command;
+}
 function printHelp() {
-    console.log(`
+    console.log(helpText());
+}
+export function helpText() {
+    return `
 @almyty/models v${VERSION}
 
 A model is usable when its card is active, has something that can call it,
-and has one passed validation run. Nothing else makes it selectable, so
+and is checked: its provider's key check passed (every model of a provider at
+once), or, for an endpoint with no provider, its own check. So
 \`list\` and \`get\` report which of those is missing rather than a name alone.
 
 Usage:
@@ -84,23 +101,23 @@ Catalog:
   list [--selectable] [--status active|inactive|error|deploying]
        [--tier public|private_cloud|local] [--provider <providerId>]
                                        List cards; each line says selectable, or why not
-  get <id>                             One card in full: capabilities, pricing, validation run
+  get <id>                             One card in full: capabilities, pricing, last check
   register --name <n> --provider <providerId> --model <vendorModelId>
            [--tier public|private_cloud|local] [--region <r>] [--context <n>]
-                                       Register a card against a stored LLM provider
-  register-endpoint --name <n> --url <baseUrl> --model <vendorModelId>
-           [--api-key-stdin] [--tier <t>] [--region <r>] [--context <n>]
-                                       Register any OpenAI-compatible server you run.
-                                       The key is prompted, or read with --api-key-stdin.
+                                       Register a card against a stored LLM provider.
+                                       A server you run (vLLM, TGI, llama.cpp) is a
+                                       \`custom\` LLM provider; register against that.
   set <id> [--name <n>] [--tier <t>] [--region <r>] [--context <n>]
            [--status active|inactive|error|deploying]
            [--price-in <usdPerMTok> --price-out <usdPerMTok>] [--clear-price]
                                        Change a card. A price pair is an override that
                                        wins over the automatic feed; --clear-price drops it.
-  sync [providerId]                    Import what a provider lists as unvalidated cards.
+  sync [providerId]                    Import what a provider lists. Selectable at once when
+                                       the provider's key check has passed.
                                        With no id, every active provider of the organization.
-  validate <id>                        One real short call. Passing is what makes a card
-                                       selectable. Exits non-zero when it fails.
+  validate <id>                        One real short call through one model. A provider's
+                                       key check covers its models; use this for an endpoint.
+                                       Exits non-zero when it fails.
   delete <id>                          Remove a card
 
 Routing (nothing is called; this plans):
@@ -118,14 +135,16 @@ Versions (optional: register an artifact only for lineage and evals on it):
                                        hf://org/repo@sha | s3://bucket/key@etag
                                        gs://bucket/key@gen | file:///path@sha
 
-Deployments:
+Hosting:
   adapters                             Registered adapters: what each can run (modelSchemes),
                                        its capabilities, and which config fields are secret
-  deploy <model> --adapter <key> [--base <b>] [--config-file <path>] [--config-stdin]
-         [--desired '<json>'] [--credential <connectionId>] [--budget <id>] [--card <cardId>]
-  deploy --model-version <id> --adapter <key> [...]
+  host <model> --adapter <key> [--base <b>] [--config-file <path>] [--config-stdin]
+       [--desired '<json>'] [--credential <connectionId>] [--budget <id>] [--card <cardId>]
+  host --model-version <id> --adapter <key> [...]
                                        <model> is where the model lives:
-                                         hf://org/repo@sha       a Hugging Face repository
+                                         hf://org/repo[@rev]     a Hugging Face repository; a
+                                                                 branch, tag or nothing is
+                                                                 pinned to its commit for you
                                          s3://bucket/prefix@etag, gs://bucket/prefix@gen,
                                          file:///path@sha
                                          bedrock:// sagemaker:// vertex:// foundry://
@@ -133,12 +152,12 @@ Deployments:
                                                                  a model already on that platform
                                        \`adapters\` lists what each provider accepts; one that
                                        cannot read your source is refused before anything runs.
-                                       Prefer --credential (a connection made with
-                                       \`almyty connections connect\`) over pasting a key.
-  deployments                          List deployments: desired vs actual, state, spend
-  deployment <id>                      One deployment in full
-  scale <deploymentId> <replicas>      Set desired replicas; 0 scales to zero
-  teardown <deploymentId>              Tear the endpoint down; weights stay in the registry
+                                       Prefer --credential (a credential added with
+                                       \`almyty credentials add\`) over pasting a key.
+  hosted                               List hosted models: desired vs actual, state, spend
+  hosted <id>                          One hosted model in full
+  scale <hostedId> <replicas>          Set desired replicas; 0 scales to zero
+  teardown <hostedId>                  Tear the endpoint down; weights stay in the registry
 
 Options:
   --json                               Undecorated JSON on stdout, for scripts
@@ -152,7 +171,7 @@ Environment:
 
 Exit codes:
 ${EXIT_CODE_HELP}
-`);
+`;
 }
 function str(flags, key) {
     const v = flags[key];
@@ -249,23 +268,6 @@ export function registerBody(flags) {
         body.contextLength = context;
     return body;
 }
-export function registerEndpointBody(flags, apiKey) {
-    const body = {
-        name: need(flags, 'name'),
-        url: need(flags, 'url'),
-        vendorModelId: need(flags, 'model'),
-    };
-    if (apiKey)
-        body.apiKey = apiKey;
-    if (str(flags, 'tier'))
-        body.privacyTier = str(flags, 'tier');
-    if (str(flags, 'region'))
-        body.region = str(flags, 'region');
-    const context = num(flags, 'context');
-    if (context !== undefined)
-        body.contextLength = context;
-    return body;
-}
 /** A card update. Refuses an empty one rather than sending a PATCH that does nothing. */
 export function setBody(flags) {
     const body = {};
@@ -336,7 +338,7 @@ export function routePolicy(flags) {
 }
 /**
  * Naming the model is configuration, so the model reference is the
- * positional argument: `deploy hf://org/repo@sha --adapter huggingface-endpoints`.
+ * positional argument: `host hf://org/repo@sha --adapter huggingface-endpoints`.
  * `--model-version` is the other way in, for people who registered an
  * artifact to get lineage and evaluation history with it.
  */
@@ -344,7 +346,7 @@ export function deployBody(flags, positional = [], providerConfig) {
     const model = positional[0] ?? str(flags, 'model');
     const modelVersion = str(flags, 'model-version');
     if (!model && !modelVersion) {
-        throw new UsageError('Name the model to run (deploy hf://org/repo@sha --adapter <key>), or pass --model-version <id>.');
+        throw new UsageError('Name the model to run (host hf://org/repo@sha --adapter <key>), or pass --model-version <id>.');
     }
     const body = { providerType: need(flags, 'adapter') };
     if (modelVersion)
@@ -394,8 +396,14 @@ export function unselectableReason(c) {
         return 'nothing can call it: no provider row and no endpoint URL';
     if (c.validationStatus !== 'passed') {
         const err = c.lastValidationError ? `: ${c.lastValidationError}` : '';
-        const status = c.validationStatus ?? 'pending';
-        return `no passed validation run (${status}${err}) — run: almyty models validate ${c.id}`;
+        if (c.validationStatus === 'failed')
+            return `the provider says this model is not available${err}`;
+        // A provider's models are usable once the provider's key check passes,
+        // which is on the connection's page under Credentials; an endpoint with
+        // no provider row is checked on its own.
+        if (c.providerId)
+            return `waiting for its provider's key check${err} (check its connection under Credentials)`;
+        return `not checked yet${err} — run: almyty models validate ${c.id}`;
     }
     return 'the catalog does not consider it selectable';
 }
@@ -431,7 +439,7 @@ export function formatCardDetail(c) {
     if (c.measuredLatencyMs)
         lines.push(`  latency       p50 ${c.measuredLatencyMs.p50 ?? '?'} ms, p95 ${c.measuredLatencyMs.p95 ?? '?'} ms`);
     if (c.deploymentId)
-        lines.push(`  deployment    ${c.deploymentId}`);
+        lines.push(`  hosted as     ${c.deploymentId}`);
     if (c.modelVersionId)
         lines.push(`  version       ${c.modelVersionId}`);
     if (!c.selectable)
@@ -529,7 +537,7 @@ export function formatSync(data) {
         }
     }
     if ((data.created?.length ?? 0) === 0 && (data.retired?.length ?? 0) === 0) {
-        lines.push('', 'Cards from a sync are unvalidated. Run `almyty models validate <id>` to make one selectable.');
+        lines.push("", "A provider's models are selectable once its key check passes; nothing is needed per model.");
     }
     return lines.join('\n');
 }
@@ -554,7 +562,7 @@ export function assertStdinIsPiped(flag) {
     if (!process.stdin.isTTY)
         return;
     throw new UsageError(`${flag} reads stdin, and stdin is your terminal, so it would wait forever.\n` +
-        `  Pipe it in:  cat config.json | almyty models deploy ... ${flag}`);
+        `  Pipe it in:  cat config.json | almyty models host ... ${flag}`);
 }
 function readStdin() {
     return new Promise((resolve, reject) => {
@@ -565,50 +573,10 @@ function readStdin() {
         process.stdin.on('error', reject);
     });
 }
-function askHidden(label) {
-    return new Promise((resolve) => {
-        const rl = createInterface({ input: process.stdin, output: process.stdout, terminal: true });
-        const anyRl = rl;
-        anyRl._writeToOutput = (s) => {
-            if (s.includes(label))
-                anyRl.output.write(label);
-        };
-        rl.question(label, (answer) => {
-            rl.close();
-            process.stdout.write('\n');
-            resolve(answer.trim());
-        });
-    });
-}
-/**
- * The endpoint key, from the safest place it can come from. `--api-key` is
- * refused because argv is world-readable; `-` means stdin, which is what a
- * script should use.
- */
-async function endpointApiKey(flags) {
-    const inline = str(flags, 'api-key');
-    if (inline && inline !== '-') {
-        throw new UsageError('--api-key puts the key in your shell history and in `ps`.\n' +
-            '  Leave it off and the key is prompted without echo, or read it from stdin:\n' +
-            '    --api-key-stdin            (also: --api-key -)\n' +
-            '  An endpoint with no key at all: --api-key ""');
-    }
-    if (flags['api-key-stdin'] || inline === '-') {
-        assertStdinIsPiped('--api-key-stdin');
-        return (await readStdin()).trim() || undefined;
-    }
-    // `--api-key` with no value, or an explicitly empty one: an open endpoint.
-    if (flags['api-key'] === true || inline === '')
-        return undefined;
-    if (!process.stdin.isTTY)
-        return undefined; // unattended and none supplied: an open endpoint
-    const typed = await askHidden('API key for the endpoint (empty for none): ');
-    return typed || undefined;
-}
 const CONFIG_ALTERNATIVES = [
     '--config-file <path>       read the JSON object from a file',
     '--config-stdin             read the JSON object from stdin',
-    '--credential <id>          use a connection made with `almyty connections connect`',
+    '--credential <id>          use a credential added with `almyty credentials add`',
 ];
 /**
  * Adapter configuration, from a file or stdin. Never a secret from argv.
@@ -618,7 +586,7 @@ const CONFIG_ALTERNATIVES = [
  * than sent blind: failing open here would make an unreachable catalog the
  * way to get a secret onto the command line.
  */
-async function deployConfig(flags, adapterSchema, schemaKnown) {
+async function hostConfig(flags, adapterSchema, schemaKnown) {
     const file = str(flags, 'config-file');
     if (file)
         return parseJsonObject(readFileSync(file, 'utf8'), `--config-file ${file}`);
@@ -657,13 +625,13 @@ async function main() {
     const client = newClient();
     const q = (path, init) => client.request(path, init);
     const post = (path, body) => q(path, { method: 'POST', body: JSON.stringify(body) });
-    switch (args.command) {
+    switch (resolveCommand(args.command)) {
         case 'list': {
             const res = await q(`/models${listQuery(args.flags)}`);
             out(args, res.data, () => (res.data.length
                 ? res.data.map(formatCard).join('\n')
                 : args.flags.selectable
-                    ? 'No selectable model cards. A card becomes selectable when one validation run passes: almyty models validate <id>'
+                    ? "No selectable models. A provider's models become selectable once its key check passes: connect one on the Models page"
                     : 'No model cards yet. Register one (almyty models register) or import a provider\'s list (almyty models sync).'));
             return;
         }
@@ -687,12 +655,6 @@ async function main() {
             out(args, res.data, () => `Registered.\n${formatCard(res.data)}\nRun: almyty models validate ${res.data.id}`);
             return;
         }
-        case 'register-endpoint': {
-            const apiKey = await endpointApiKey(args.flags);
-            const res = await post('/models/register-endpoint', registerEndpointBody(args.flags, apiKey));
-            out(args, res.data, () => `Registered.\n${formatCard(res.data)}\nRun: almyty models validate ${res.data.id}`);
-            return;
-        }
         case 'set': {
             const id = needArg(args.positional, 0, 'card id', 'set <id> [--tier t] [--price-in n --price-out n] ...');
             const res = await q(`/models/${id}`, { method: 'PATCH', body: JSON.stringify(setBody(args.flags)) });
@@ -736,7 +698,7 @@ async function main() {
             out(args, res.data, () => (res.data.length ? res.data.map(formatAdapter).join('\n') : 'No adapters registered.'));
             return;
         }
-        case 'deploy': {
+        case 'host': {
             // The adapter's schema says which config fields are secret, so the
             // check happens before anything is sent.
             const adapterKey = need(args.flags, 'adapter');
@@ -755,27 +717,27 @@ async function main() {
                 if (err instanceof UsageError)
                     throw err;
                 // The catalog could not be read. The API still validates the body,
-                // but --config can no longer be screened, so deployConfig refuses it.
+                // but --config can no longer be screened, so hostConfig refuses it.
             }
-            const providerConfig = await deployConfig(args.flags, adapterSchema, schemaKnown);
+            const providerConfig = await hostConfig(args.flags, adapterSchema, schemaKnown);
             const res = await post('/model-deployments', deployBody(args.flags, args.positional, providerConfig));
             out(args, res.data, () => `Queued. Reconcile picks it up within a couple of minutes.\n${formatDeployment(res.data)}`);
             return;
         }
-        case 'deployments': {
+        case 'hosted': {
+            const id = args.positional[0];
+            if (id) {
+                const res = await q(`/model-deployments/${id}`);
+                out(args, res.data, () => formatDeploymentDetail(res.data));
+                return;
+            }
             const res = await q('/model-deployments');
-            out(args, res.data, () => (res.data.length ? res.data.map(formatDeployment).join('\n') : 'No deployments.'));
-            return;
-        }
-        case 'deployment': {
-            const id = needArg(args.positional, 0, 'deployment id', 'deployment <id>');
-            const res = await q(`/model-deployments/${id}`);
-            out(args, res.data, () => formatDeploymentDetail(res.data));
+            out(args, res.data, () => (res.data.length ? res.data.map(formatDeployment).join('\n') : 'No hosted models.'));
             return;
         }
         case 'scale': {
-            const id = needArg(args.positional, 0, 'deployment id', 'scale <deploymentId> <replicas>');
-            const raw = needArg(args.positional, 1, 'replica count', 'scale <deploymentId> <replicas>');
+            const id = needArg(args.positional, 0, 'hosted model id', 'scale <hostedId> <replicas>');
+            const raw = needArg(args.positional, 1, 'replica count', 'scale <hostedId> <replicas>');
             const replicas = Number(raw);
             if (!Number.isInteger(replicas) || replicas < 0)
                 throw new UsageError(`replicas must be a whole number of zero or more, got ${raw}`);
@@ -784,7 +746,7 @@ async function main() {
             return;
         }
         case 'teardown': {
-            const id = needArg(args.positional, 0, 'deployment id', 'teardown <deploymentId>');
+            const id = needArg(args.positional, 0, 'hosted model id', 'teardown <hostedId>');
             const res = await post(`/model-deployments/${id}/teardown`, {});
             out(args, res.data, () => formatDeployment(res.data));
             return;
@@ -795,8 +757,23 @@ async function main() {
             process.exit(EXIT.USAGE);
     }
 }
-const invokedDirectly = process.argv[1] && /models-cli|almyty-models|dist\/index\.js|src\/index\.ts/.test(process.argv[1]) && !process.env.VITEST;
-if (invokedDirectly) {
+/**
+ * True when `argv1` is this module, directly or through the npm bin
+ * symlink. The old check matched the path against a regex and bailed out
+ * when VITEST was set, so a smoke test that spawned the built CLI from
+ * vitest inherited VITEST and got no output at all.
+ */
+export function isEntrypoint(argv1, moduleUrl) {
+    if (!argv1)
+        return false;
+    try {
+        return realpathSync(argv1) === realpathSync(fileURLToPath(moduleUrl));
+    }
+    catch {
+        return false;
+    }
+}
+if (isEntrypoint(process.argv[1], import.meta.url)) {
     main().catch((err) => {
         console.error(describeError(err, process.env.ALMYTY_URL));
         process.exit(exitCodeFor(err));

package/package.json

@@ -1,10 +1,10 @@
 {
   "name": "@almyty/models",
-  "version": "1.5.0",
+  "version": "1.5.2",
   "publishConfig": {
     "access": "public"
   },
-  "description": "Manage the almyty model catalog from your terminal: register cards, validate them, deploy weights, watch spend.",
+  "description": "Manage the almyty model catalog from your terminal: register cards, validate them, host weights, watch spend.",
   "type": "module",
   "main": "dist/index.js",
   "bin": {
@@ -28,7 +28,7 @@
   "author": "almyty",
   "license": "Apache-2.0",
   "dependencies": {
-    "@almyty/client": "^1.2.0"
+    "@almyty/client": "^1.5.1"
   },
   "devDependencies": {
     "@types/node": "^25.4.0",