Self-hosted, Self-Improving, Federated Everything App — dev management, agent orchestration, persistent mind, digital twin, second brain, and creative pipeline from a single dashboard over Tailscale to port the user into the OS
JavaScript ai-agentsdashboarddeveloper-toolsdigital-twinquantified-selfsecond-brainself-hostedtailscale
/**
* Harnesses — manage the coding-agent CLIs/TUIs PortOS drives.
*
* GET /api/harnesses → every harness, its version, its providers
* POST /api/harnesses/action?runtime=&action= → SSE stream of install/update/remove
* POST /api/harnesses/models/refresh → re-read a harness's model catalog
*
* Backs **Models → Harnesses**. The AI Providers page could already install a
* missing CLI (`POST /api/providers/runtimes/install`), but there was no way to
* see which version was installed, update a stale one, remove one, or refresh
* the model list a harness knows about — which is how an install goes months
* out of date with nothing in the UI saying so.
*
* Every mutating endpoint names a harness *id* from the fixed table in
* `services/providerRuntimeInstaller.js`; no request value reaches a shell word,
* and no response carries a resolved filesystem path (a global bin directory
* embeds the host account name — see the Sensitive Data rules in AGENTS.md).
*/
import { Router } from 'express';
import { asyncHandler, ServerError } from '../lib/errorHandler.js';
import { validateRequest, harnessActionSchema, harnessRefreshSchema } from '../lib/validation.js';
import { streamHarnessAction } from '../services/harnessActionStream.js';
import { listHarnesses, refreshHarnessModels } from '../services/harnesses.js';
const router = Router();
/**
* The whole page in one round trip. `?fresh=1` bypasses both the runtime status
* TTL and the npm-registry cache — what the page's Refresh button sends, and
* what it re-reads after every action so a just-installed version shows without
* waiting out a cache.
*/
router.get('/', asyncHandler(async (req, res) => {
res.json({ harnesses: await listHarnesses({ fresh: req.query.fresh === '1' }) });
}));
/**
* Install, update, or remove one harness, streaming the child's output as SSE.
*
* A POST because it mutates host state, and the shared `RuntimeInstallModal`
* already appends `runtime` to the query string for every BYO-runtime
* installer — `action` rides beside it for the same reason.
*/
router.post('/action', asyncHandler(async (req, res) => {
const { runtime, action } = validateRequest(harnessActionSchema, {
runtime: req.query.runtime,
action: req.query.action,
});
await streamHarnessAction(req, res, { runtime, action });
}));
/**
* Re-read a harness's own model catalog and write it to every provider that
* draws from it.
*
* Refusals here are 409, not 500: "this harness cannot list its models" and
* "sign in first" are states of the host, not server faults, and the page
* renders the reason verbatim.
*/
router.post('/models/refresh', asyncHandler(async (req, res) => {
const { runtime } = validateRequest(harnessRefreshSchema, { runtime: req.query.runtime });
const result = await refreshHarnessModels(runtime);
if (!result.ok) {
throw new ServerError(result.reason, { status: 409, code: 'HARNESS_MODELS_UNAVAILABLE', context: { runtime } });
}
res.json(result);
}));
export default router;