Skip to content

Studio UI — Component Reference

Every screen and control in the G-1 Studio UI, and the exact API call behind it. Use this page when you want to do from code what you just did by clicking — or when you are debugging and need to know which endpoint a panel is actually hitting.

How to read the paths

The Studio UI is a single-page app served from the same origin as the API, so every path below is same-origin and relative.

Prefix Reaches Documented in
/v1/… G-1 Studio (control plane) Studio API map
/gw/… G1-Proxy (data plane), with /gw stripped before it arrives G1-Proxy API map

So the Knowledge Base page calling /gw/v1/glad/rag/collections is calling the proxy's /v1/glad/rag/collections.

The header every per-app control sends

The Application picker in the top bar writes the active app id to browser storage, and the UI attaches it as X-Geodesia-App on every proxy call that is scoped per Application — RAG, feedback, audio, web search, chat. If you reproduce those calls from code and omit the header, you get the default Application's data, not the one you were looking at.

Reproducing any control from code

Every row below is a real HTTP call. Take the path, prefix the host, and send it — the UI has no private API.

# what the Application picker reads
curl -s http://localhost:8080/v1/glad/apps | jq '[.apps[].app_id]'

# what a per-Application panel sends (note the header)
curl -s http://localhost:8080/gw/v1/glad/rag/collections \
  -H "X-Geodesia-App: support_bot" | jq

Chrome — always on screen

Component What it is Calls
Sidebar Navigation. Entries come from the router, not from the API. —
App Switcher (top bar) Picks the active Application. Populated from the application store. Selecting one re-scopes every page below. GET /v1/glad/apps
App Scope Banner The strip reminding you which Application you are looking at. Reads the store; makes no call of its own. —
Version readout Both component versions side by side — they are released independently and can differ. Available to the UI through a shared composable; not every build surfaces it in the chrome. GET /v1/glad/version (Studio) · GET /gw/version (Proxy)
Update / licence banner Polls for an available update and for a licence expiring within 14 days. GET /gw/v1/glad/gateway/notifications
Backend-ready gate Blocks the UI with a "starting up" state until both services answer. GET /v1/glad/health · GET /gw/v1/glad/gateway/calibration
Calibration banner The strip that appears while a closed-book recalibration is running. GET /gw/v1/glad/gateway/calibration · GET /v1/glad/calibration/status

Dashboard

Route / — the landing page.

Control What it shows Calls
Health tiles Overall compliance posture. GET /v1/glad/dashboard
Regulatory scorecard Per-framework coverage. GET /v1/glad/scorecard
Traffic & detection charts Passed / blocked / hallucinated / unsafe over time, scoped to the active Application. GET /v1/glad/dashboard/charts?application_id=…

Applications

Route /applications. The control-plane editor. See Managing Applications.

Control What it does Calls
Application list GET /v1/glad/apps
New Application POST /v1/glad/apps 🔑
Name / config save PUT /v1/glad/apps/{app_id} 🔑
Delete DELETE /v1/glad/apps/{app_id} 🔑
Pause / Resume Stop or resume traffic without losing the configuration. POST /v1/glad/apps/{app_id}/pause · /resume 🔑
Kill Per-Application kill switch. POST /v1/glad/apps/{app_id}/kill 🔑
Axis threshold sliders + enforcement selectors The nine axes, each with a threshold and block / annotate / off. The set of sliders is built from what the served detector reports it has. GET /v1/glad/apps/meta → PUT /v1/glad/apps/{app_id}/policy 🔑
Upstream panel (type, base URL, model, key) field edits → PUT /v1/glad/apps/{app_id} 🔑
Discover models button Lists what the backend offers. Sends api_key: "***" plus application_id so the stored key is used and the browser never holds it. POST /v1/glad/apps/upstream/models
Test connection Connects, lists models, probes log-probability support, returns a sample reply and its latency. POST /gw/upstream/test
Test capabilities Runs the detector on built-in safe/unsafe pairs and reports which axes separate. Touches no upstream. POST /gw/test-capabilities 🔒
Complex routing panel Enable, threshold, and the Model-B binding. PUT /v1/glad/apps/{app_id}/routing 🔑
Cost panel Token rates, monthly budget, alert bands, over-budget policy. PUT /v1/glad/apps/{app_id}/cost 🔑
Cost preview GET …/cost/summary · …/cost/daily · …/cost/forecast
API keys table List, mint, revoke. The plaintext key is shown once, in the create response. GET/POST /v1/glad/apps/{app_id}/keys · DELETE …/keys/{key_id} 🔑
Export data Downloads everything this Application saw. Honours the server's Content-Disposition filename. GET /v1/glad/apps/{app_id}/export?fmt=sqlite\|csv\|jsonl 🔑
Model calibration card Embedded per-app closed-book recalibration for the selected model. GET /v1/glad/calibration/capabilities · /status · POST /v1/glad/calibration/run

Playground

Route /chat (the screen is called Playground in the sidebar — it is not there to chat, it is there to try: you send a turn and watch the nine axes, the thresholds and the certificate on what comes back). The interactive console — every detection surface in one place.

Control What it does Calls
Message composer → Send Sends the turn through the full pipeline. Attaches the active Application, the enforcement mode, threshold overrides, the thinking level, RAG selection and the web-search flag as body fields. POST /gw/v1/chat/completions (proxy path) or POST /v1/chat/completions (local-model path)
Thinking level dropdown Sets thinking_level 0–3 on the request body. Level 3 is MAX. body field on the chat call — Thinking Levels
Mode toggle (block / passthrough) Sets mode on the body, overriding the Application's enforcement for this turn only. body field — Enforcement Modes
Web search toggle Sets web_search: true. body field — Live Web Search
Knowledge base picker Sets rag: {collection_id, …}. body field — Knowledge Base
Microphone button Streams PCM to the voice guard and brakes mid-utterance. WS /gw/v1/glad/audio/stream, config from GET /gw/v1/glad/audio/status
Score overlay / Energy panel Per-axis probability, threshold and flag. Read straight from geodesia.axis_energy in the chat response — no second call. —
Token heatmap / Prompt XAI / Response XAI cards The dual-surface attribution that runs automatically at verdict time: which prompt tokens caused a block, which answer tokens caused a flag. POST /gw/v1/glad/causal-explainability/analyze with method: "dca_dual"
RAG sources panel Retrieved passages and claim verification. From geodesia.rag in the response. —
Web research panel Pages searched, screened and used. From the response. —
Flag this answer Files a correction into the review queue. The form's axes and verdicts come from the schema endpoint rather than being hard-coded. GET /gw/v1/glad/feedback/schema → POST /gw/v1/glad/feedback
Bank badge How many approved corrections are live in the exemplar bank. GET /gw/v1/glad/feedback/bank/status

Knowledge Base

Route /knowledge. See Knowledge Base. Everything here is on the proxy, scoped by X-Geodesia-App.

Control What it does Calls
Collection list GET /gw/v1/glad/rag/collections
New collection POST /gw/v1/glad/rag/collections
Delete collection DELETE /gw/v1/glad/rag/collections/{id}
Upload document Multipart POST returns 202 immediately; the UI then polls the progress endpoint until stage is done or error. It does not hold the connection open — a large PDF would outlive any proxy timeout. POST …/collections/{id}/documents → poll GET /gw/v1/glad/rag/ingest/progress
Delete document DELETE …/collections/{id}/documents/{doc_id}
Test query box Retrieve passages without sending a chat turn. POST …/collections/{id}/query
Status strip Whether the RAG stack is available at all. GET /gw/v1/glad/rag/status

Cost & FinOps

Route /cost. See Cost & FinOps.

Control What it shows Calls
Organisation total GET /v1/glad/orgs/{org_id}/cost/summary
Per-Application spend GET /v1/glad/apps/{app_id}/cost/summary
Daily series chart GET /v1/glad/apps/{app_id}/cost/daily
Budget & projection chart Month-end projection against the budget band. GET /v1/glad/apps/{app_id}/cost/forecast?method=run_rate

FRIA

Route /fria. See FRIA.

Control What it does Calls
Dossier list GET /v1/glad/fria?deployer_id=…
New FRIA POST /v1/glad/fria
Auto-prefill Drafts the answers the deployment can already prove. GET /v1/glad/fria/auto-prefill
Save PUT /v1/glad/fria/{id}
Approve POST /v1/glad/fria/{id}/approve
Archive POST /v1/glad/fria/{id}/archive
Runtime evidence panel The measured half of the dossier: what the last 90 days of real traffic actually shows. GET /v1/glad/fria/{id}/runtime-evidence?window_days=90
Export (PDF / DOCX / JSON) Runs as a background job with a progress state, because a full render can outlive a connection timeout. POST /v1/glad/fria/{id}/export/async → GET /v1/glad/doc-jobs/{job_id}/status?token=… → …/download?token=…

Kill Switch

Route /kill-switch. See Kill Switch.

Control What it does Calls
State indicator GET /v1/glad/kill-switch/status?deployer_id=…
Activate (reason required) POST /v1/glad/kill-switch/activate
Deactivate POST /v1/glad/kill-switch/deactivate

Oversight

Route /oversight. See Human Oversight.

Control What it does Calls
Review queue Pending reviews, scoped to the active Application. GET /v1/glad/oversight/pending?application_id=…
Counters GET /v1/glad/oversight/summary
Approve / Reject / Escalate / Modify Sends review_id plus the decision. escalated opens a fresh review one level up. POST /v1/glad/oversight/decide
Verify chain button Proves the decision trail has not been altered. GET /v1/glad/chain/verify
Download all chats (JSONL) button Exports every stored message of the selected Application — one JSON object per line, ordered by (session, time). Downloading another Application's transcripts would be handing one customer another's data, so the scope is the same one that governs the rest of the screen. Against a backend without the route it falls back to assembling in the browser and says so (compatibility mode), because that path is N+1 requests and not the same thing. GET /v1/glad/chat-export?application_id=…

Feedback

Route /feedback. The curation queue. See Human Feedback Loop. All on the proxy.

Control What it does Calls
Queue table Filterable by status, axis, region, Application. GET /gw/v1/glad/feedback?status=…&axis=…
Counters GET /gw/v1/glad/feedback/stats
Approve / Reject with axis, verdict, note, weight POST /gw/v1/glad/feedback/{id}/review
Delete DELETE /gw/v1/glad/feedback/{id}
Export corpus GET /gw/v1/glad/feedback/export?status=approved
Bank status GET /gw/v1/glad/feedback/bank/status
Retrain — memory Refreshes the exemplar bank immediately. POST /gw/v1/glad/feedback/retrain {mode:"memory"}
Retrain — weights Exports a corpus and launches the trainer; the button then polls the job. POST …/retrain {mode:"weights"} → GET …/retrain/status?job_id=…
Auto-judge card The idle-CPU reviewer: on/off, model, queue depths, disagreements, promotions. GET /gw/v1/glad/feedback/auto/status · /auto/config · PUT /auto/config · GET /auto/items
Preview judge prompt Read-only. You configure the description of your organisation; the audit protocol around it is not a setting. GET /gw/v1/glad/feedback/auto/prompt-preview?axis=…

Audit & Chain

Route /audit. See Audit Chain and AI Watermark.

Control What it does Calls
Ledger snapshot + recent entries GET /v1/glad/chain/status
Retention status GET /v1/glad/retention/status
Verify watermark Paste text, or give call_id + watermark_id for a full live verification. POST /v1/glad/watermark/verify

Reports

Route /reports. See Reports & Manuals.

Control What it does Calls
Generate report (data) POST /v1/glad/report
Download report PDF Background job with progress. POST /v1/glad/report/pdf/async → doc-job poll → download
Deployer manual (data) POST /v1/glad/deployer-manual
Download manual PDF Background job with progress. POST /v1/glad/deployer-manual/pdf/async → doc-job poll → download

If a render fails, the failure reason surfaces from the job's error field — it is the message that tells the user what to fix (no logged calls, no applicable law selected, …), not a generic failure.

Route /legal.

Control Calls
Framework list GET /v1/glad/legal/frameworks
Framework detail GET /v1/glad/legal/{framework_id}

Causal Explainability

Route /causal-explainability. See Causal Explainability.

Control What it does Calls
Session / message picker Loads a past turn to analyse. GET /v1/glad/chat-sessions · GET /v1/glad/chat-messages?session_id=…
Analyse — proxy path Black-box attribution over the detector. No local generator, no GPU. This is the path an OpenAI-API deployment uses. POST /gw/v1/glad/causal-explainability/analyze
Analyse — local path Needs a locally loaded model. Runs as a job with a progress bar. POST /v1/glad/causal-explainability/jobs → GET …/jobs/{job_id}
Cancel Aborts the in-flight request, not just the poll loop. client-side abort
Dual-surface cards Prompt and answer attributed separately. analyze with method: "dca_dual"

MuPAX

Route /mupax.

Control Calls
Explain POST /v1/glad/mupax/explain
Language selector GET /v1/glad/mupax/languages

Model Calibration

Route /calibration.

Control Calls
Capability check GET /v1/glad/calibration/capabilities
Artifact registry GET /v1/glad/calibration/registry
Probe generator POST /v1/glad/calibration/probe
Run (fast / deep) POST /v1/glad/calibration/run
Live status GET /v1/glad/calibration/status
Stop POST /v1/glad/calibration/stop

Settings

Route /settings. See Settings.

Card What it configures Calls
Licence Paste a vendor-signed key; read tier, company, daily chat limit, models, expiry. Applying or removing is immediate. GET/POST/DELETE /gw/v1/glad/gateway/license · GET /gw/v1/glad/gateway/entitlements
Gateway Upstream binding, enforcement defaults, constitutional prompt switch, streaming cadence, numeric solver. Saves apply live. GET/POST /gw/v1/glad/gateway/config
PII guard status Whether redaction is effective — an enabled switch on an image without the library protects nothing, and this card says so instead of showing a green shield. GET /gw/v1/glad/pii/status
MCP Which MCP servers and interceptors are listening, and the per-surface actions. Read-only here; configured by env. GET /gw/v1/glad/mcp/status
Voice input (ASR) Whisper model picker for the live-voice guard, plus a Test microphone button. GET /gw/v1/glad/audio/status · POST /gw/v1/glad/audio/utterance
Web search Store the provider API key out-of-band. The key is never returned in clear — only a masked hint. env_locked means an environment variable wins. GET/POST /gw/v1/glad/websearch/config
Database SQLite by default; point Studio at an existing PostgreSQL, test it, deploy the schema. GEODESIA_DB_URL always overrides; a switch takes effect after a restart. GET /v1/glad/db/config · POST /v1/glad/db/test · /db/deploy · /db/use-sqlite
Detection thresholds (deployer profile) The deployer-wide default profile, distinct from per-Application policy. GET/POST /v1/glad/threshold-prefs

Gateway settings talk to the proxy directly when it is remote

When the configured gateway is co-located (localhost, 127.0.0.1, ::1) or the URL is unparseable, the UI goes through the same-origin /gw proxy. Only a genuinely remote gateway is addressed by its own URL. If you are reproducing these calls, target the proxy directly.

Documentation & API Docs

Route What it is Calls
/documentation The bundled user guide, rendered in-product so it works air-gapped. GET /v1/glad/documentation (Studio) or GET /gw/v1/glad/documentation (proxy-only deployments)
/api-docs An in-product, browsable endpoint catalogue. Static data compiled into the UI — it makes no API call to build the list. —

Customer Tokens

Route /_g1-license-vault — unlisted, not in the sidebar. Requires GLAD_LICENSE_ADMIN_KEY supplied as X-Geodesia-Admin-Key; without the variable configured server-side the endpoints return 503.

Control Calls
Model scope list GET /v1/glad/license-tokens/models
Issue token POST /v1/glad/license-tokens 🔐

Listing and revoking are API-only — GET /v1/glad/license-tokens and POST /v1/glad/license-tokens/{token_id}/revoke, both 🔐. This page only issues.