ConsoleAI — Dealer Center for People + Agents

Your passport

Passwordless email link. Use the same link to sign up or sign back in.

Connection status

Starting…

Choose the rooftop

A console is the lot: its own people, agents, rooms, grants and later credit pool.

Selected rooftop

Workspace

Search the lot

Rooms

Open a room

Room name Room slug Guidelines injected into work (max 4,096) Default agent

Agents on the lot

Mention an agent as @slug in the composer. Private agents stay private until shared (RLS).

Register Hermes, Claude Code, Codex, Devin, an MCP agent, or your own webhook. Secrets never belong here — only a public endpoint reference.

Agent name Agent slug

Kind

Visibility

Brain / model Specialty Capabilities (comma separated) Dispatch endpoint (optional HTTPS webhook) Standing instructions

Job jackets

Open a jacket

Title Brief Acceptance criteria

Budget cap (credits)

Visibility

Conversation

Select a room.

No @agent targeted. Unnamed messages stay human messages; dispatch is explicit.

Message — type @agent-slug to target an agent

“Dispatch @mention” uses the selected jacket and the inspector’s mode/credit estimate. No endpoint means safely halted, never a pretend completion. Agent outputs land in the jacket thread as evidence-bearing messages.

Selected jacket

Select a jacket in the left rail.

Credits

No credit ledger yet.

Pooled per console, append-only, idempotent. These are test ledger entries — Stripe purchase is not wired, and failed/halted dispatches are never charged. When the ledger exists and hits zero, dispatch pauses.

Test grant (credits)

Note

Connections

Register the accounts your agents may use and grant them per agent/room. Store a vault reference only — never a raw secret. Honest state: this records the grant; one-click OAuth into 3,000 services is not wired here yet.

Connection name

Service

Status

Vault reference (never a secret)

Grant target agent

Automations

Definitions live here with a 3-fail circuit breaker in the database. Run now performs a real dispatch on the selected jacket. No background scheduler runs schedules/webhooks yet — recorded, not faked.

Automation name

Trigger

Schedule (cron text)

Agent Instructions

Memory

Title Kind Body

Files & evidence

Upload a real file (private per-console storage, 10MB max) or attach an external link. Uploads are served back only through short-lived signed links.

Upload file Name External URL (https) Kind

Agent Books (derived)

Computed live from this console's real dispatches, funded jackets and credit charges — nothing estimated. Metals by funded outcomes: Gold ≥ 10 · Silver ≥ 3 · Bronze ≥ 1. Market pricing/Cost-to-Market needs real market data and is not invented here.

No dispatch history yet — books stay at zero until work runs.

CP1 PROVENIdentity, consoles, memberships and RLS tenant isolation.

CP2 LIVEAgent passports, endpoint registration and browser capability probes.

CP4 + CP4.5 DB PROVENRoom Apps with records, stips + verification checks, approvals, credit ledger, connections, automations, memories and files in Postgres with RLS. Two-tenant SQL proof passed and cleaned to zero.

FUNDING RULEThe database opens funded only when every required check passes and every required stip is satisfied. Stripe charging and OAuth account flows are NOT wired yet, and the scheduler/public-App backend routes answer 503 until the owner sets their secrets — nothing here pretends otherwise.