mcoda 0.1.76 → 0.1.79

package/CHANGELOG.md

@@ -4,6 +4,10 @@
 - Initial npm packaging scaffold for the mcoda CLI.
 - Added bundled mswarm consent terms plus guided `mcoda setup`/postinstall consent bootstrap for installed CLI packages.
 
+## 0.1.78
+- Added owner-local GPU job commands: `mcoda gpu list`, `mcoda gpu ops`, and GPU-aware `mcoda job artifact upload|run|status|logs|events|artifacts|cancel|retry`.
+- Added README and usage docs for the generic GPU job connection flags and environment fallbacks.
+
 ## 0.1.76
 - Added top-level help aliases: `mcoda help`, `mcoda --help`, `mcoda -h`, and `mcoda -H`.
 - Release v0.1.76.

package/README.md

@@ -34,6 +34,8 @@ mcoda docs pdr generate --workspace-root . --project WEB --rfp-path docs/rfp/web
 - Execution: `mcoda add-tests`, `mcoda work-on-tasks`, `mcoda code-review`, `mcoda qa-tasks`
 - Backlog: `mcoda backlog`, `mcoda task`
 - Jobs/telemetry: `mcoda jobs`, `mcoda tokens`, `mcoda telemetry`
+- Owner-local GPU jobs: `mcoda gpu list`, `mcoda gpu ops`, `mcoda job artifact upload|run|status|logs|events|artifacts|cancel|retry`
+- Self-hosted agents: `mcoda self-hosted agent list`, `mcoda self-hosted agent details`, `mcoda self-hosted agent sync`
 - Agents: `mcoda test-agent`, `mcoda agent-run`
 - Updates: `mcoda update --check`
 
@@ -45,9 +47,22 @@ If that sufficiency pass errors, create-tasks continues (fail-open) and records
 Environment variables are optional overrides for workspace settings:
 - `MCODA_DOCDEX_URL` to point at a docdex server.
 - `MCODA_API_BASE_URL` or `MCODA_JOBS_API_URL` for job APIs.
+- `MCODA_MSWARM_NODE_BASE_URL`, `MCODA_MSWARM_NODE_ID`, and `MCODA_MSWARM_NODE_SIGNING_SECRET` for owner-local generic GPU job commands.
 - `MCODA_TELEMETRY` set to `off` to disable telemetry.
 - `MCODA_STREAM_IO=1` to emit agent I/O lines to stderr.
 
+## Self-hosted mswarm agents
+
+`mcoda self-hosted agent list` reads direct self-hosted agents and opt-in
+load-balanced aliases from mswarm when the configured API key allows it. Direct
+entries stay pinned to one server. Load-balanced aliases are saved as auto
+routes and let mswarm choose an eligible upgraded node for the requested model
+or capability.
+
+Use direct slugs for rollback or pinned-server workloads. Use auto aliases only
+when the product should allow mswarm to route around busy, drained, stale, or
+incompatible nodes.
+
 ## Programmatic usage
 ```ts
 import { McodaEntrypoint } from "mcoda";

package/dist/bin/McodaEntrypoint.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"McodaEntrypoint.d.ts","sourceRoot":"","sources":["../../src/bin/McodaEntrypoint.ts"],"names":[],"mappings":";AAuDA,qBAAa,eAAe;WACb,GAAG,CAAC,IAAI,GAAE,MAAM,EAA0B,GAAG,OAAO,CAAC,IAAI,CAAC;CAoPxE"}
+{"version":3,"file":"McodaEntrypoint.d.ts","sourceRoot":"","sources":["../../src/bin/McodaEntrypoint.ts"],"names":[],"mappings":";AAyDA,qBAAa,eAAe;WACb,GAAG,CAAC,IAAI,GAAE,MAAM,EAA0B,GAAG,OAAO,CAAC,IAAI,CAAC;CAwPxE"}

package/dist/bin/McodaEntrypoint.js

@@ -6,6 +6,7 @@ import { AgentsCommands } from '../commands/agents/AgentsCommands.js';
 import { CloudCommands } from '../commands/cloud/CloudCommands.js';
 import { SelfHostedCommands } from '../commands/self-hosted/SelfHostedCommands.js';
 import { WorkersCommands } from '../commands/workers/WorkersCommands.js';
+import { GpuCommands } from '../commands/gpu/GpuCommands.js';
 import { ConfigCommands } from '../commands/config/ConfigCommands.js';
 import { ConsentCommands } from '../commands/consent/ConsentCommands.js';
 import { GatewayAgentCommand } from '../commands/agents/GatewayAgentCommand.js';
@@ -35,7 +36,7 @@ import { AgentRunCommand } from '../commands/agents/AgentRunCommand.js';
 import { SetWorkspaceCommand } from '../commands/workspace/SetWorkspaceCommand.js';
 import { ProjectGuidanceCommand } from '../commands/workspace/ProjectGuidanceCommand.js';
 import { MswarmConfigStore } from '@mcoda/core';
-const TOP_LEVEL_USAGE = 'Usage: mcoda <agent|cloud|cloud-agent|self-hosted|self-hosted-agent|workers|worker|config|consent|setup|gateway-agent|test-agent|agent-run|routing|docs|openapi|job|jobs|tokens|telemetry|create-tasks|migrate-tasks|refine-tasks|task-sufficiency-audit|sds-preflight|order-tasks|tasks|add-tests|work-on-tasks|gateway-trio|code-review|qa-tasks|backlog|task|task-detail|estimate|update|set-workspace|project-guidance|pdr|sds> [...args]\n' +
+const TOP_LEVEL_USAGE = 'Usage: mcoda <agent|cloud|cloud-agent|self-hosted|self-hosted-agent|workers|worker|gpu|config|consent|setup|gateway-agent|test-agent|agent-run|routing|docs|openapi|job|jobs|tokens|telemetry|create-tasks|migrate-tasks|refine-tasks|task-sufficiency-audit|sds-preflight|order-tasks|tasks|add-tests|work-on-tasks|gateway-trio|code-review|qa-tasks|backlog|task|task-detail|estimate|update|set-workspace|project-guidance|pdr|sds> [...args]\n' +
     'Setup: use `mcoda setup` after installation (or accept the postinstall prompt) to complete the mandatory mswarm telemetry consent flow.\n' +
     'Config: use `mcoda config set mswarm-api-key <KEY>` to persist an encrypted mswarm API key in the resolved global mcoda config file.\n' +
     'Consent: use `mcoda consent accept` before other commands if you need to complete consent outside the guided setup flow.\n' +
@@ -43,10 +44,11 @@ const TOP_LEVEL_USAGE = 'Usage: mcoda <agent|cloud|cloud-agent|self-hosted|self-
     'Cloud agents: use `mcoda cloud agent list|details|sync` to discover and materialize mswarm-managed remote agents.\n' +
     'Self-hosted agents: use `mcoda self-hosted agent list|details|sync` to discover and materialize owner-hosted mswarm agents.\n' +
     'Workers: use `mcoda workers list|details|sync|run` to discover, materialize, and invoke mswarm Workers.\n' +
+    'GPU jobs: use `mcoda gpu list` and `mcoda job artifact upload|run|status|logs|events|artifacts|cancel` for owner-local generic GPU jobs.\n' +
     'Expose this machine: install `@mcoda/mswarm`, then run `mswarm install <KEY>`.\n' +
     'Aliases: `tasks order-by-deps` forwards to `order-tasks` (dependency-aware ordering), `task`/`task-detail` show a single task.\n' +
     'Help: use `mcoda help`, `mcoda --help`, `mcoda -h`, or `mcoda -H` for this overview.\n' +
-    'Job commands (mcoda job --help for details): list|status|watch|logs|inspect|resume|cancel|tokens\n' +
+    'Job commands (mcoda job --help for details): list|status|watch|logs|inspect|resume|cancel|tokens plus GPU job artifact|run|events|artifacts with node options\n' +
     'Jobs API required for job commands (set MCODA_API_BASE_URL/MCODA_JOBS_API_URL or workspace api.baseUrl). status/watch/logs exit non-zero on failed/cancelled jobs per SDS.';
 const TOP_LEVEL_HELP_COMMANDS = new Set(['--help', '-h', '-H', 'help']);
 export class McodaEntrypoint {
@@ -135,6 +137,10 @@ export class McodaEntrypoint {
             await WorkersCommands.run(rest);
             return;
         }
+        if (command === 'gpu') {
+            await GpuCommands.run(rest);
+            return;
+        }
         if (command === 'config') {
             await ConfigCommands.run(rest);
             return;

package/dist/commands/gpu/GpuCommands.d.ts

@@ -0,0 +1,8 @@
+export declare class GpuCommands {
+    static run(argv: string[]): Promise<void>;
+}
+export declare class GpuJobCommands {
+    static shouldHandle(argv: string[]): boolean;
+    static run(argv: string[]): Promise<void>;
+}
+//# sourceMappingURL=GpuCommands.d.ts.map

package/dist/commands/gpu/GpuCommands.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"GpuCommands.d.ts","sourceRoot":"","sources":["../../../src/commands/gpu/GpuCommands.ts"],"names":[],"mappings":"AAoSA,qBAAa,WAAW;WACT,GAAG,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,OAAO,CAAC,IAAI,CAAC;CAiChD;AAED,qBAAa,cAAc;IACzB,MAAM,CAAC,YAAY,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,OAAO;WAQ/B,GAAG,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,OAAO,CAAC,IAAI,CAAC;CA8FhD"}

package/dist/commands/gpu/GpuCommands.js

@@ -0,0 +1,382 @@
+import { createHash, randomUUID } from "node:crypto";
+import { basename } from "node:path";
+import { readFile } from "node:fs/promises";
+import YAML from "yaml";
+import { MswarmApi } from "@mcoda/core";
+const GPU_USAGE = `
+Usage: mcoda gpu <list|ops> [options]
+
+Commands:
+  gpu list                 List owner-local GPU/job capability projection
+  gpu ops                  Show owner-local generic job queue, usage, quota, and audit summary
+
+Connection options:
+  --node-base-url <URL>    Owner-local node URL (or MCODA_MSWARM_NODE_BASE_URL)
+  --node-id <ID>           Node id for signing capability requests
+  --signing-secret <KEY>   Owner-local generic job signing secret
+  --token <TOKEN>          Pre-signed capability or ops token
+  --timeout-ms <N>         Request timeout
+  --audit-limit <N>        Audit rows for gpu ops (default 50, max 250)
+  --audit-offset <N>       Audit row offset for gpu ops
+  --json                   Emit JSON
+`.trim();
+const JOB_USAGE = `
+Usage: mcoda job <artifact|run|status|logs|events|artifacts|cancel|retry> [options]
+
+GPU job commands:
+  job artifact upload <FILE>
+    --job-id <ID> --request-id <ID> --node-id <ID> --job-type <TYPE>
+    [--artifact-path <PATH>] [--artifact-name <NAME>] [--content-type <TYPE>]
+  job run --job-file <FILE> [--wait] [--json]
+  job run --type <TYPE> --payload-file <FILE> [--wait] [--json]
+  job status <JOB_ID> [--job-file <FILE> | --node-id ... --request-id ... --job-type ...]
+  job logs <JOB_ID>
+  job events <JOB_ID>
+  job artifacts <JOB_ID>
+  job cancel <JOB_ID>
+  job retry <JOB_ID>
+
+Connection options:
+  --node-base-url <URL>    Owner-local node URL (or MCODA_MSWARM_NODE_BASE_URL)
+  --node-id <ID>           Node id for signing requests
+  --signing-secret <KEY>   Owner-local generic job signing secret
+  --token <TOKEN>          Pre-signed generic job token
+  --timeout-ms <N>         Request timeout
+  --json                   Emit JSON
+`.trim();
+const GPU_JOB_SUBCOMMANDS = new Set(["artifact", "run", "events", "artifacts"]);
+const SHARED_JOB_SUBCOMMANDS = new Set(["status", "logs", "cancel", "retry"]);
+const GPU_JOB_FLAGS = new Set([
+    "--gpu",
+    "--node-base-url",
+    "--node-id",
+    "--signing-secret",
+    "--token",
+    "--job-file",
+    "--payload-file",
+    "--job-type",
+    "--type",
+    "--schema-version",
+    "--request-id"
+]);
+const TERMINAL_STATES = new Set(["succeeded", "failed", "cancelled", "expired", "blocked"]);
+const parseArgs = (argv) => {
+    const flags = {};
+    const positionals = [];
+    for (let index = 0; index < argv.length; index += 1) {
+        const arg = argv[index];
+        if (arg.startsWith("--")) {
+            const key = arg.replace(/^--/, "");
+            const next = argv[index + 1];
+            if (next && !next.startsWith("--")) {
+                const current = flags[key];
+                if (current === undefined)
+                    flags[key] = next;
+                else if (Array.isArray(current))
+                    flags[key] = [...current, next];
+                else if (typeof current === "string")
+                    flags[key] = [current, next];
+                else
+                    flags[key] = [next];
+                index += 1;
+            }
+            else {
+                flags[key] = true;
+            }
+            continue;
+        }
+        positionals.push(arg);
+    }
+    return { flags, positionals };
+};
+const resolveString = (value) => {
+    if (value === undefined || typeof value === "boolean")
+        return undefined;
+    return Array.isArray(value) ? value[value.length - 1] : value;
+};
+const resolvePositiveInt = (value, label) => {
+    const raw = resolveString(value);
+    if (raw === undefined)
+        return undefined;
+    const parsed = Number.parseInt(raw, 10);
+    if (!Number.isFinite(parsed) || parsed <= 0) {
+        throw new Error(`Invalid ${label}; expected a positive integer`);
+    }
+    return parsed;
+};
+const resolveNonNegativeInt = (value, label) => {
+    const raw = resolveString(value);
+    if (raw === undefined)
+        return undefined;
+    const parsed = Number.parseInt(raw, 10);
+    if (!Number.isFinite(parsed) || parsed < 0) {
+        throw new Error(`Invalid ${label}; expected a non-negative integer`);
+    }
+    return parsed;
+};
+const nodeBaseUrl = (parsed) => resolveString(parsed.flags["node-base-url"]) || process.env.MCODA_MSWARM_NODE_BASE_URL;
+const nodeId = (parsed) => resolveString(parsed.flags["node-id"]) || process.env.MCODA_MSWARM_NODE_ID || process.env.MSWARM_SELF_HOSTED_NODE_ID;
+const signingSecret = (parsed) => resolveString(parsed.flags["signing-secret"]) ||
+    process.env.MCODA_MSWARM_NODE_SIGNING_SECRET ||
+    process.env.MSWARM_SELF_HOSTED_INVOCATION_SIGNING_SECRET;
+const baseAuth = (parsed) => ({
+    nodeBaseUrl: nodeBaseUrl(parsed),
+    token: resolveString(parsed.flags.token),
+    signingSecret: signingSecret(parsed),
+    tokenTtlSeconds: resolvePositiveInt(parsed.flags["token-ttl-seconds"], "--token-ttl-seconds")
+});
+const createApi = async (parsed) => MswarmApi.create({
+    baseUrl: nodeBaseUrl(parsed),
+    apiKey: resolveString(parsed.flags["api-key"]),
+    timeoutMs: resolvePositiveInt(parsed.flags["timeout-ms"], "--timeout-ms")
+});
+const readStructuredFile = async (file) => {
+    const raw = await readFile(file, "utf8");
+    if (file.endsWith(".yaml") || file.endsWith(".yml")) {
+        return YAML.parse(raw);
+    }
+    return JSON.parse(raw);
+};
+const asRecord = (value) => typeof value === "object" && value !== null && !Array.isArray(value)
+    ? value
+    : {};
+const readJobEnvelope = async (parsed) => {
+    const file = resolveString(parsed.flags["job-file"]) || resolveString(parsed.flags["payload-file"]);
+    if (!file) {
+        throw new Error("GPU job run requires --job-file or --payload-file");
+    }
+    const payload = asRecord(await readStructuredFile(file));
+    if (typeof payload.job_id === "string" && typeof payload.request_id === "string" && typeof payload.node_id === "string") {
+        return payload;
+    }
+    const schemaVersion = resolveString(parsed.flags["schema-version"]) ||
+        (typeof payload.schema_version === "string" ? payload.schema_version : "2026-06-14");
+    const jobType = resolveString(parsed.flags.type) ||
+        resolveString(parsed.flags["job-type"]) ||
+        (typeof payload.job_type === "string" ? payload.job_type : undefined);
+    if (!jobType) {
+        throw new Error("--job-file/--payload-file must contain job_type or use --type/--job-type");
+    }
+    return {
+        job_id: resolveString(parsed.flags["job-id"]) || `job-${randomUUID()}`,
+        request_id: resolveString(parsed.flags["request-id"]) || `req-${randomUUID()}`,
+        node_id: requireText(nodeId(parsed), "--node-id"),
+        job: {
+            ...payload,
+            schema_version: schemaVersion,
+            job_type: jobType
+        }
+    };
+};
+const referenceFromArgs = async (parsed, explicitJobId) => {
+    const jobFile = resolveString(parsed.flags["job-file"]) || resolveString(parsed.flags["payload-file"]);
+    if (jobFile) {
+        const job = await readJobEnvelope(parsed);
+        return {
+            ...baseAuth(parsed),
+            jobId: explicitJobId || job.job_id,
+            nodeId: job.node_id,
+            requestId: job.request_id,
+            schemaVersion: job.job.schema_version,
+            jobType: job.job.job_type
+        };
+    }
+    return {
+        ...baseAuth(parsed),
+        jobId: requireText(explicitJobId || resolveString(parsed.flags["job-id"]), "JOB_ID"),
+        nodeId: nodeId(parsed),
+        requestId: resolveString(parsed.flags["request-id"]),
+        schemaVersion: resolveString(parsed.flags["schema-version"]) || "2026-06-14",
+        jobType: resolveString(parsed.flags["job-type"]) || resolveString(parsed.flags.type)
+    };
+};
+function requireText(value, label) {
+    if (!value || !value.trim()) {
+        throw new Error(`${label} is required`);
+    }
+    return value.trim();
+}
+const printJsonOrValue = (value, json) => {
+    // eslint-disable-next-line no-console
+    console.log(json ? JSON.stringify(value, null, 2) : formatValue(value));
+};
+const formatValue = (value) => {
+    if (typeof value === "string")
+        return value;
+    return JSON.stringify(value, null, 2);
+};
+const printSnapshot = (snapshot, json) => {
+    if (json) {
+        printJsonOrValue(snapshot, true);
+        return;
+    }
+    const job = snapshot.job;
+    // eslint-disable-next-line no-console
+    console.log(`${job.job_id} ${job.state}`);
+    if (job.backpressure?.message) {
+        // eslint-disable-next-line no-console
+        console.log(job.backpressure.message);
+    }
+    if (job.result?.error?.message) {
+        // eslint-disable-next-line no-console
+        console.log(job.result.error.message);
+    }
+};
+const formatOpsSummary = (summary) => {
+    const lines = [
+        `${summary.node.node_id} ${summary.node.generic_jobs_enabled ? "generic-jobs-enabled" : "generic-jobs-disabled"}`,
+        `queue active=${summary.queue.active_jobs} queued=${summary.queue.queued_jobs} terminal=${summary.queue.terminal_jobs}`,
+        `quota available=${summary.quota.available_slots}/${summary.quota.max_concurrent_jobs} production_enforced=${summary.quota.production_enforced}`,
+        `usage total=${summary.usage.total_jobs} succeeded=${summary.usage.succeeded_jobs} failed=${summary.usage.failed_jobs} cancelled=${summary.usage.cancelled_jobs} blocked=${summary.usage.blocked_jobs} gpu_seconds=${summary.usage.gpu_seconds}`,
+    ];
+    if (summary.queue.jobs.length) {
+        lines.push("jobs:");
+        for (const job of summary.queue.jobs.slice(0, 10)) {
+            lines.push(`  ${job.job_id} ${job.state} ${job.job_type} tenant=${job.tenant_id}`);
+        }
+    }
+    if (summary.audit.events.length) {
+        lines.push(`audit offset=${summary.audit.offset} limit=${summary.audit.limit} total=${summary.audit.total}:`);
+        for (const event of summary.audit.events.slice(0, 10)) {
+            lines.push(`  ${event.timestamp} ${event.action} ${event.job_id}`);
+        }
+    }
+    return lines.join("\n");
+};
+const waitForJob = async (api, reference, intervalMs) => {
+    let snapshot = await api.getGenericJob(reference);
+    while (!TERMINAL_STATES.has(snapshot.job.state)) {
+        await new Promise((resolve) => setTimeout(resolve, intervalMs));
+        snapshot = await api.getGenericJob(reference);
+    }
+    return snapshot;
+};
+export class GpuCommands {
+    static async run(argv) {
+        const [subcommand, ...rest] = argv;
+        if (!subcommand || argv.includes("--help") || argv.includes("-h")) {
+            // eslint-disable-next-line no-console
+            console.log(GPU_USAGE);
+            return;
+        }
+        if (subcommand !== "list" && subcommand !== "ops") {
+            throw new Error(`Unknown gpu subcommand: ${subcommand}`);
+        }
+        const parsed = parseArgs(rest);
+        const api = await createApi(parsed);
+        try {
+            if (subcommand === "list") {
+                const capabilities = await api.listGpuCapabilities({
+                    ...baseAuth(parsed),
+                    nodeId: nodeId(parsed)
+                });
+                printJsonOrValue(capabilities, Boolean(parsed.flags.json));
+                return;
+            }
+            const ops = await api.getGenericJobOps({
+                ...baseAuth(parsed),
+                nodeId: nodeId(parsed),
+                auditLimit: resolvePositiveInt(parsed.flags["audit-limit"], "--audit-limit"),
+                auditOffset: resolveNonNegativeInt(parsed.flags["audit-offset"], "--audit-offset")
+            });
+            // eslint-disable-next-line no-console
+            console.log(parsed.flags.json ? JSON.stringify(ops, null, 2) : formatOpsSummary(ops));
+        }
+        finally {
+            await api.close();
+        }
+    }
+}
+export class GpuJobCommands {
+    static shouldHandle(argv) {
+        const [subcommand] = argv;
+        if (!subcommand)
+            return false;
+        if (GPU_JOB_SUBCOMMANDS.has(subcommand))
+            return true;
+        if (!SHARED_JOB_SUBCOMMANDS.has(subcommand))
+            return false;
+        return argv.some((arg) => GPU_JOB_FLAGS.has(arg) || arg.startsWith("--node-base-url="));
+    }
+    static async run(argv) {
+        const [subcommand, ...rest] = argv;
+        if (!subcommand || argv.includes("--help") || argv.includes("-h")) {
+            // eslint-disable-next-line no-console
+            console.log(JOB_USAGE);
+            return;
+        }
+        const parsed = parseArgs(rest);
+        const api = await createApi(parsed);
+        try {
+            switch (subcommand) {
+                case "artifact": {
+                    const action = parsed.positionals[0];
+                    const file = parsed.positionals[1];
+                    if (action !== "upload" || !file) {
+                        throw new Error("Usage: mcoda job artifact upload <FILE> [options]");
+                    }
+                    const content = await readFile(file);
+                    const artifactPath = resolveString(parsed.flags["artifact-path"]) || basename(file);
+                    const result = await api.uploadGenericJobArtifact({
+                        ...(await referenceFromArgs(parsed, resolveString(parsed.flags["job-id"]))),
+                        name: resolveString(parsed.flags["artifact-name"]) || basename(file),
+                        path: artifactPath,
+                        contentBase64: content.toString("base64"),
+                        contentType: resolveString(parsed.flags["content-type"]),
+                        sha256: createHash("sha256").update(content).digest("hex"),
+                        sizeBytes: content.length
+                    });
+                    printJsonOrValue(result, Boolean(parsed.flags.json));
+                    break;
+                }
+                case "run": {
+                    const job = await readJobEnvelope(parsed);
+                    const snapshot = await api.runGenericJob(job, baseAuth(parsed));
+                    const finalSnapshot = parsed.flags.wait
+                        ? await waitForJob(api, await referenceFromArgs(parsed, job.job_id), resolvePositiveInt(parsed.flags["interval-ms"], "--interval-ms") || 1000)
+                        : snapshot;
+                    printSnapshot(finalSnapshot, Boolean(parsed.flags.json));
+                    break;
+                }
+                case "status": {
+                    const snapshot = await api.getGenericJob(await referenceFromArgs(parsed, parsed.positionals[0]));
+                    printSnapshot(snapshot, Boolean(parsed.flags.json));
+                    break;
+                }
+                case "logs": {
+                    const result = await api.getGenericJobLogs(await referenceFromArgs(parsed, parsed.positionals[0]));
+                    if (parsed.flags.json)
+                        printJsonOrValue(result, true);
+                    else
+                        printJsonOrValue(result.logs.map((log) => log.message).join("\n"), false);
+                    break;
+                }
+                case "events": {
+                    const result = await api.getGenericJobEvents(await referenceFromArgs(parsed, parsed.positionals[0]));
+                    printJsonOrValue(result, Boolean(parsed.flags.json));
+                    break;
+                }
+                case "artifacts": {
+                    const result = await api.getGenericJobArtifacts(await referenceFromArgs(parsed, parsed.positionals[0]));
+                    printJsonOrValue(result, Boolean(parsed.flags.json));
+                    break;
+                }
+                case "cancel": {
+                    const result = await api.cancelGenericJob(await referenceFromArgs(parsed, parsed.positionals[0]));
+                    printSnapshot(result, Boolean(parsed.flags.json));
+                    break;
+                }
+                case "retry": {
+                    const result = await api.retryGenericJob(await referenceFromArgs(parsed, parsed.positionals[0]));
+                    printSnapshot(result, Boolean(parsed.flags.json));
+                    break;
+                }
+                default:
+                    throw new Error(`Unknown GPU job subcommand: ${subcommand}`);
+            }
+        }
+        finally {
+            await api.close();
+        }
+    }
+}

package/dist/commands/jobs/JobsCommands.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"JobsCommands.d.ts","sourceRoot":"","sources":["../../../src/commands/jobs/JobsCommands.ts"],"names":[],"mappings":"AACA,OAAO,EAAsE,aAAa,EAA0C,MAAM,aAAa,CAAC;AAGxJ,KAAK,aAAa,GAAG,MAAM,GAAG,QAAQ,GAAG,OAAO,GAAG,MAAM,GAAG,SAAS,GAAG,QAAQ,GAAG,QAAQ,GAAG,QAAQ,CAAC;AAEvG,MAAM,WAAW,aAAa;IAC5B,UAAU,EAAE,aAAa,CAAC;IAC1B,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,OAAO,CAAC;IACf,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,WAAW,CAAC,EAAE,OAAO,CAAC;IACtB,KAAK,CAAC,EAAE,OAAO,CAAC;CACjB;AAuCD,eAAO,MAAM,YAAY,GAAI,MAAM,MAAM,EAAE,KAAG,aAyH7C,CAAC;AA8BF,eAAO,MAAM,mBAAmB,GAAI,UAAU,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,KAAG,MAAM,GAAG,SAiBhF,CAAC;AAEF,eAAO,MAAM,kBAAkB,GAAI,UAAU,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,KAAG,MAAM,GAAG,SAsB/E,CAAC;AAaF,eAAO,MAAM,eAAe,GAAI,MAAM,aAAa,EAAE,KAAG,IA8CvD,CAAC;AAwJF,qBAAa,YAAY;WACV,GAAG,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,OAAO,CAAC,IAAI,CAAC;CA6OhD"}
+{"version":3,"file":"JobsCommands.d.ts","sourceRoot":"","sources":["../../../src/commands/jobs/JobsCommands.ts"],"names":[],"mappings":"AACA,OAAO,EAAsE,aAAa,EAA0C,MAAM,aAAa,CAAC;AAIxJ,KAAK,aAAa,GAAG,MAAM,GAAG,QAAQ,GAAG,OAAO,GAAG,MAAM,GAAG,SAAS,GAAG,QAAQ,GAAG,QAAQ,GAAG,QAAQ,CAAC;AAEvG,MAAM,WAAW,aAAa;IAC5B,UAAU,EAAE,aAAa,CAAC;IAC1B,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,OAAO,CAAC;IACf,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,WAAW,CAAC,EAAE,OAAO,CAAC;IACtB,KAAK,CAAC,EAAE,OAAO,CAAC;CACjB;AAuCD,eAAO,MAAM,YAAY,GAAI,MAAM,MAAM,EAAE,KAAG,aAyH7C,CAAC;AA8BF,eAAO,MAAM,mBAAmB,GAAI,UAAU,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,KAAG,MAAM,GAAG,SAiBhF,CAAC;AAEF,eAAO,MAAM,kBAAkB,GAAI,UAAU,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,KAAG,MAAM,GAAG,SAsB/E,CAAC;AAaF,eAAO,MAAM,eAAe,GAAI,MAAM,aAAa,EAAE,KAAG,IA8CvD,CAAC;AAwJF,qBAAa,YAAY;WACV,GAAG,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,OAAO,CAAC,IAAI,CAAC;CAiPhD"}

package/dist/commands/jobs/JobsCommands.js

@@ -1,5 +1,6 @@
 import path from "node:path";
 import { JobInsightsService, JobResumeService, JobService, TelemetryService, WorkspaceResolver } from "@mcoda/core";
+import { GpuJobCommands } from "../gpu/GpuCommands.js";
 const usage = `mcoda job <list|status|watch|logs|inspect|resume|cancel|tokens> ...
 
 Commands:
@@ -401,6 +402,10 @@ const handleTokens = async (parsed, workspace) => {
 };
 export class JobsCommands {
     static async run(argv) {
+        if (GpuJobCommands.shouldHandle(argv)) {
+            await GpuJobCommands.run(argv);
+            return;
+        }
         const parsed = parseJobArgs(argv);
         const workspace = await WorkspaceResolver.resolveWorkspace({
             cwd: process.cwd(),

package/dist/commands/self-hosted/SelfHostedCommands.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"SelfHostedCommands.d.ts","sourceRoot":"","sources":["../../../src/commands/self-hosted/SelfHostedCommands.ts"],"names":[],"mappings":"AA6NA,qBAAa,kBAAkB;WAChB,GAAG,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,OAAO,CAAC,IAAI,CAAC;CA8FhD"}
+{"version":3,"file":"SelfHostedCommands.d.ts","sourceRoot":"","sources":["../../../src/commands/self-hosted/SelfHostedCommands.ts"],"names":[],"mappings":"AAyOA,qBAAa,kBAAkB;WAChB,GAAG,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,OAAO,CAAC,IAAI,CAAC;CAkGhD"}

package/dist/commands/self-hosted/SelfHostedCommands.js

@@ -7,6 +7,7 @@ Subcommands:
     --provider <NAME>       Filter by provider (mcoda|ollama)
     --limit <N>             Limit returned agents
     --include-unreachable   Include unreachable agents in the catalog result
+    --include-load-balanced Include auto-routed load-balanced self-hosted aliases
     --max-cost-per-1m-token <N>
                              Exclude agents above the given cost_per_million
     --sorted-by-catalog-rating
@@ -14,9 +15,11 @@ Subcommands:
     --min-context <N>       Require at least this context window
     --min-reasoning <N>     Require at least this reasoning rating
   agent details <SLUG>      Show a single mswarm self-hosted agent (supports --json)
+    --include-load-balanced Allow details for auto-routed aliases
   agent sync                Sync self-hosted agents into the local mcoda registry
     --provider <NAME>       Filter by provider before syncing
     --include-unreachable   Sync unreachable agents too
+    --include-load-balanced Sync auto-routed load-balanced aliases too
     --limit <N>             Limit synced agents
     --prune                 Remove previously synced self-hosted agents missing from the current catalog result
     --agent-slug-prefix <P> Override the local managed-agent slug prefix
@@ -104,6 +107,7 @@ const resolveNonNegativeNumber = (value, label) => {
 const formatNumber = (value) => value === undefined || Number.isNaN(value) ? "-" : String(value);
 const formatCapabilities = (capabilities) => capabilities && capabilities.length > 0 ? capabilities.join(",") : "-";
 const formatBoolean = (value) => value === undefined ? "-" : value ? "yes" : "no";
+const agentRoutingMode = (agent) => agent.load_balanced ? "auto" : "direct";
 const pad = (value, width) => value.padEnd(width, " ");
 const renderTable = (headers, rows) => {
     const widths = headers.map((header, columnIndex) => Math.max(header.length, ...rows.map((row) => row[columnIndex]?.length ?? 0)));
@@ -122,6 +126,7 @@ const printAgentList = (agents) => {
     }
     const headers = [
         "REMOTE SLUG",
+        "ROUTE",
         "PROVIDER",
         "ADAPTER",
         "MODEL",
@@ -136,6 +141,7 @@ const printAgentList = (agents) => {
     ];
     const rows = agents.map((agent) => [
         agent.remote_slug ?? agent.slug,
+        agentRoutingMode(agent),
         agent.provider,
         agent.adapter ?? "-",
         agent.default_model,
@@ -155,6 +161,8 @@ const printAgentDetails = (agent) => {
     const entries = [
         ["Slug", agent.slug],
         ["Remote slug", agent.remote_slug ?? "-"],
+        ["Route", agentRoutingMode(agent)],
+        ["Load-balanced group", agent.load_balanced_group_id ?? "-"],
         ["Provider", agent.provider],
         ["Adapter", agent.adapter ?? "-"],
         ["Source agent", agent.source_agent_slug ?? "-"],
@@ -170,6 +178,7 @@ const printAgentDetails = (agent) => {
         ["Supports tools", formatBoolean(agent.supports_tools)],
         ["Supports reasoning", formatBoolean(agent.supports_reasoning)],
         ["Health", agent.health_status ?? "-"],
+        ["Members", formatNumber(agent.member_count)],
         ["Capabilities", formatCapabilities(agent.capabilities)],
     ];
     const labelWidth = Math.max(...entries.map(([label]) => label.length));
@@ -187,11 +196,12 @@ const printSyncSummary = (summary) => {
         record.remoteSlug,
         record.localSlug,
         record.action,
+        record.routingMode ?? "-",
         record.provider,
         record.defaultModel,
     ]);
     // eslint-disable-next-line no-console
-    console.log(renderTable(["REMOTE SLUG", "LOCAL SLUG", "ACTION", "PROVIDER", "MODEL"], rows));
+    console.log(renderTable(["REMOTE SLUG", "LOCAL SLUG", "ACTION", "ROUTE", "PROVIDER", "MODEL"], rows));
 };
 export class SelfHostedCommands {
     static async run(argv) {
@@ -227,6 +237,7 @@ export class SelfHostedCommands {
                         provider: resolveString(parsed.flags.provider),
                         limit: resolvePositiveInt(parsed.flags.limit, "--limit"),
                         includeUnreachable: Boolean(parsed.flags["include-unreachable"]),
+                        includeLoadBalanced: Boolean(parsed.flags["include-load-balanced"]),
                         maxCostPerMillion: resolveNonNegativeNumber(parsed.flags["max-cost-per-1m-token"], "--max-cost-per-1m-token"),
                         minContextWindow: resolvePositiveInt(parsed.flags["min-context"], "--min-context"),
                         minReasoningRating: resolveNonNegativeNumber(parsed.flags["min-reasoning"], "--min-reasoning"),
@@ -246,7 +257,9 @@ export class SelfHostedCommands {
                     if (!slug) {
                         throw new Error("Usage: mcoda self-hosted agent details <SLUG> [--json]");
                     }
-                    const agent = await api.getSelfHostedAgent(slug);
+                    const agent = await api.getSelfHostedAgent(slug, {
+                        includeLoadBalanced: Boolean(parsed.flags["include-load-balanced"]),
+                    });
                     if (parsed.flags.json) {
                         // eslint-disable-next-line no-console
                         console.log(JSON.stringify(agent, null, 2));
@@ -261,6 +274,7 @@ export class SelfHostedCommands {
                         provider: resolveString(parsed.flags.provider),
                         limit: resolvePositiveInt(parsed.flags.limit, "--limit"),
                         includeUnreachable: Boolean(parsed.flags["include-unreachable"]),
+                        includeLoadBalanced: Boolean(parsed.flags["include-load-balanced"]),
                         pruneMissing: Boolean(parsed.flags.prune),
                     });
                     if (parsed.flags.json) {

package/dist/index.d.ts

@@ -3,6 +3,7 @@ export * from './bin/McodaEntrypoint.js';
 export * from './commands/cloud/CloudCommands.js';
 export * from './commands/self-hosted/SelfHostedCommands.js';
 export * from './commands/workers/WorkersCommands.js';
+export * from './commands/gpu/GpuCommands.js';
 export * from './commands/config/ConfigCommands.js';
 export * from './commands/consent/ConsentCommands.js';
 export * from './commands/docs/DocsCommands.js';

package/dist/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,cAAc,qCAAqC,CAAC;AACpD,cAAc,0BAA0B,CAAC;AACzC,cAAc,mCAAmC,CAAC;AAClD,cAAc,8CAA8C,CAAC;AAC7D,cAAc,uCAAuC,CAAC;AACtD,cAAc,qCAAqC,CAAC;AACpD,cAAc,uCAAuC,CAAC;AACtD,cAAc,iCAAiC,CAAC;AAChD,cAAc,iCAAiC,CAAC;AAChD,cAAc,uCAAuC,CAAC;AACtD,cAAc,2CAA2C,CAAC;AAC1D,cAAc,uCAAuC,CAAC;AACtD,cAAc,wCAAwC,CAAC;AACvD,cAAc,yCAAyC,CAAC;AACxD,cAAc,yCAAyC,CAAC;AACxD,cAAc,2CAA2C,CAAC;AAC1D,cAAc,2CAA2C,CAAC;AAC1D,cAAc,oDAAoD,CAAC;AACnE,cAAc,4CAA4C,CAAC;AAC3D,cAAc,uCAAuC,CAAC;AACtD,cAAc,uCAAuC,CAAC;AACtD,cAAc,wCAAwC,CAAC;AACvD,cAAc,uCAAuC,CAAC;AACtD,cAAc,qCAAqC,CAAC;AACpD,cAAc,uCAAuC,CAAC;AACtD,cAAc,kCAAkC,CAAC;AACjD,cAAc,uCAAuC,CAAC;AACtD,cAAc,gDAAgD,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,cAAc,qCAAqC,CAAC;AACpD,cAAc,0BAA0B,CAAC;AACzC,cAAc,mCAAmC,CAAC;AAClD,cAAc,8CAA8C,CAAC;AAC7D,cAAc,uCAAuC,CAAC;AACtD,cAAc,+BAA+B,CAAC;AAC9C,cAAc,qCAAqC,CAAC;AACpD,cAAc,uCAAuC,CAAC;AACtD,cAAc,iCAAiC,CAAC;AAChD,cAAc,iCAAiC,CAAC;AAChD,cAAc,uCAAuC,CAAC;AACtD,cAAc,2CAA2C,CAAC;AAC1D,cAAc,uCAAuC,CAAC;AACtD,cAAc,wCAAwC,CAAC;AACvD,cAAc,yCAAyC,CAAC;AACxD,cAAc,yCAAyC,CAAC;AACxD,cAAc,2CAA2C,CAAC;AAC1D,cAAc,2CAA2C,CAAC;AAC1D,cAAc,oDAAoD,CAAC;AACnE,cAAc,4CAA4C,CAAC;AAC3D,cAAc,uCAAuC,CAAC;AACtD,cAAc,uCAAuC,CAAC;AACtD,cAAc,wCAAwC,CAAC;AACvD,cAAc,uCAAuC,CAAC;AACtD,cAAc,qCAAqC,CAAC;AACpD,cAAc,uCAAuC,CAAC;AACtD,cAAc,kCAAkC,CAAC;AACjD,cAAc,uCAAuC,CAAC;AACtD,cAAc,gDAAgD,CAAC"}

package/dist/index.js

@@ -3,6 +3,7 @@ export * from './bin/McodaEntrypoint.js';
 export * from './commands/cloud/CloudCommands.js';
 export * from './commands/self-hosted/SelfHostedCommands.js';
 export * from './commands/workers/WorkersCommands.js';
+export * from './commands/gpu/GpuCommands.js';
 export * from './commands/config/ConfigCommands.js';
 export * from './commands/consent/ConsentCommands.js';
 export * from './commands/docs/DocsCommands.js';

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "mcoda",
-  "version": "0.1.76",
+  "version": "0.1.79",
   "description": "Local-first CLI for planning, documentation, and execution workflows with agent assistance.",
   "type": "module",
   "main": "dist/index.js",
@@ -47,12 +47,12 @@
   },
   "dependencies": {
     "yaml": "^2.4.2",
-    "@mcoda/shared": "0.1.76",
-    "@mcoda/core": "0.1.76"
+    "@mcoda/core": "0.1.79",
+    "@mcoda/shared": "0.1.79"
   },
   "devDependencies": {
-    "@mcoda/db": "0.1.76",
-    "@mcoda/integrations": "0.1.76"
+    "@mcoda/db": "0.1.79",
+    "@mcoda/integrations": "0.1.79"
   },
   "scripts": {
     "build": "tsc -p tsconfig.json",