CDB HUB.
--:--:--
Live · Orchestration Layer

Un orchestre.
dix agents.

CDB HUB coordonne 10 agents spécialisés — routing every task to gpt-5.5 or MiniMax-M3 by complexity.

● gateway telegram Aucun run actif
Recent · Activity
LIVE5m1h24h7d

Agents that
ship work.

    — ACTIVE MISSIONS
    0
    open on the board —
    CPU — · RAM — live
    — FOCUS

    The agent carrying the largest share of the workload

    0
    % OF TASKS
    — TASKS RESOLVED · BY AGENT

    All logged runs

    0 —
    completed / total
    — FLEET · 10 AGENTS

    Per-agent state & throughput

      — BUILD

      Spin up
      a new agent

      / click + above

      — INFERENCE LEDGER · MODEL ROUTING

      Tasks routed by complexity · real runs

      — models
      0 tokens
      across all logged runs
      Offloaded to fast model0%
      Complexity routing — → gpt-5.5 · — → MiniMax-M3
      ModelTasks / share
        — CDB HUB / Flotte IA

        Pilotage
        intelligent
        10 agents.

        just start running / agents

        Hermes Core · v4.2.1 Channel · Secure Operator 01 · Tier IV
        Fleet · Configuration

        Dix agents.
        un cockpit.

        Inspect each specialist, route them to the right model, and watch the heatbeat of the entire fleet in one place.

        0 tasks logged
        — avg success
        0% routed to fast
        — TASK SUMMARY · 24H

        Throughput across the specialist fleet

        Total tasks
        802
        ↗ +12% vs yesterday
        Avg success
        98.9%
        5-agent rolling avg
          — MODEL ROUTING

          Route each agent
          to its best brain.

          5 agents · 7 models

          / changes apply on next task dispatch

          — ACTIVITY HEATMAP · 7d × 24h

          When the fleet is busiest, UTC

          cool hot
          Peak hour
          14:00 UTC
          Quietest
          03:00 UTC
          Weekday avg
          74%
          — TASK DISTRIBUTION

          Share of total cycles

            — AGENT LOGS · RECENT

            Every logged run across the fleet

            0
            TimeAgentTaskModelStatus
              Hermes Core · v4.2.1 Channel · Secure Operator 01 · Tier IV
              Mission Board · 10 active

              Every mission, in motion.

              Drag a card between columns to move it through your workflow. This is your personal task board — capture what you're working on, set priorities, and track it to done.

              To do
              4
              In progress
              3
              Done
              3
              The Empire · Live skyline

              A city built by agents.

              Every specialist owns a tower. The orchestrator runs HQ at the center. Lit windows mean live work. Drag to orbit, scroll to zoom, click a building to open its dossier.

              Buildings
              5
              Lights on
              0
              Tasks
              0
              Booting the empire… First visit compiles the 3D engine (≈10s). If this lingers, hard-reload the page.
              — AGENT OUTPUT

              Library.

              Total Docs
              0
              Agents Writing
              0
              Latest
              —
                Select a document to read
                — AUTOMATION

                Schedule.

                Hermes Jobs
                0
                Active
                0
                Next Run
                —
                — HERMES JOBS

                Agent-driven scheduled tasks · run, pause, resume or delete

                — DIRECT LINE

                Talk to the fleet.

                idle
                Documentation · How this dashboard works

                Know every corner.

                A plain-language guide to every tab, panel, and number on your Hermes Mission Control dashboard — what it shows, where the data comes from, and how to use it. Nothing here is invented: every figure traces back to your real Hermes activity or to your own local files.

                Tab · Overview

                Your fleet at a glance

                The landing screen — one live snapshot of the whole system: health, load, output, and cost. It refreshes on its own every few seconds, so what you see is always current.

                Gateway & Telegram status
                The pills at the top show whether your Hermes gateway is running and whether Telegram is connected. They flip the moment either changes. source · gateway_state.json
                Active Missions
                How many tasks are live on your board right now (To do + In progress) and how many are already done — a quick read on your workload. source · board.db
                VPS line (CPU · RAM · Disk)
                Live health of the server the dashboard runs on, sampled directly on the machine — no guesswork. If a number climbs, you know before it becomes a problem. source · /proc + statvfs
                Focus ring
                A single gauge for how concentrated the fleet's recent work is — a fast "how loaded is the system" read without digging into per-agent numbers.
                Tasks Resolved · by agent
                Total completed agent runs and the overall success rate. This is real activity — every row is an actual agent run that was logged. source · agent-logs.db
                Inference Ledger · Model Routing
                Total tokens used and, if you've enabled the optional routing add-on, the split between your premium and fast models plus the "offload %" (the share sent to the cheaper/faster model). Without routing it simply shows which models your agents actually used. It never shows a made-up "money saved" figure. source · agent-logs.db (+ optional model-routing.json)
                Activity ticker
                A live scroll of the most recent things your agents did — the newest actions first, updating as work happens. source · agent-logs.db
                Tab · Agents

                Every specialist, measured

                The performance view for your five agents — the Orchestrator plus Scout, Scribe, Reach, and Dev. See who's carrying the load, how reliable each one is, and exactly what they've been doing.

                Fleet header stats
                Top-line totals for the last 24 hours: tasks logged, average success rate, and the share of work routed to your fast model.
                Agent cards
                One card per agent showing its role, its most recent task, its default model, task count, success rate, and a load bar. Click any card to open its detail drawer.
                Agent detail drawer
                Slides in from the right with one agent's full record: total tasks, success rate, which models it ran on and how often, its most recent tasks, and its last error (if any). Close it with ✕, the Esc key, or by clicking outside. source · agent-logs.db
                Task Distribution
                How the total workload is split across the five agents — at a glance, who is doing most of the work.
                Agent Logs · Recent
                The raw feed at the bottom of the tab — every logged run as a row: time, agent, task, model, and status (completed / failed). This is the source of truth behind the headline numbers. source · agent-logs.db
                Tab · Chat

                A direct line to the fleet

                Talk to any agent right from the dashboard. Every message runs a real agent turn — not a canned reply — threaded into that agent's live session with full memory.

                The roster
                Your five agents down the left. Pick one to open its conversation. Each keeps its own memory, so it remembers what you told it before.
                The Orchestrator = your main agent
                First in the roster and marked ● telegram thread. It's your default Hermes agent — the one Telegram routes to. When you message it here, both your message and its reply also appear in your Telegram, so the dashboard and your phone stay in sync.
                The specialists
                Scout, Scribe, Reach, and Dev each resume their own private session with full memory. They don't have a Telegram thread — they're worked from the dashboard.
                Good to know
                Sending a message spends tokens because it runs a live turn on your model. History is read straight from your Hermes session; the send goes through your real Hermes agent. source · Hermes state.db (read) + hermes CLI (send)
                Tab · Tasks

                Your personal mission board

                This board is yours — your own to-do list, not agent work. Three columns move a task from idea to done. It comes seeded with a couple of example cards per column that you can edit or delete.

                The three columns
                To do (not started), In progress (working on it), and Done (completed). The counts in the hero update as you move cards.
                New Mission
                Add a task with a title, a priority (P1 Critical / P2 High / P3 Normal), and a starting column. It saves instantly and stays after a refresh.
                Drag & drop
                Drag a card between columns to change its status — the move is saved automatically.
                Search & priority filter
                Find a task by title or id, or filter the board down to a single priority (ALL / P1 / P2 / P3).
                The card & where it lives
                Each card shows an id, a priority badge (its colour also tints the left edge), tags, a progress bar, and a delete button that appears on hover. Tasks are stored in the dashboard's own local database — completely separate from your Hermes data. source · board.db
                Tab · Office

                Your fleet as a living city

                A 3D skyline where each agent is a building and the HQ tower sits at the centre. It's a fun, glance-able way to see who's working right now. (The 3D view needs a GPU / WebGL-capable browser.)

                Blue vs. orange
                A tower glows blue when its agent is actively running a turn, and orange when idle. Message an agent from the Chat tab and watch its tower light up. source · live working set from the server
                The counters
                The small readouts show how many buildings there are, how many have their lights on (agents currently working), and the day's task count.
                Tab · Content

                Everything your agents write

                A living library of the long-form work your agents produce — articles, research briefs, scripts, plans. Each agent saves to its own folder, and it all shows up here to read, edit, download, or delete.

                Stats strip
                Total documents, how many agents are actively writing, and the most recent piece.
                Reading & filtering
                Click any document to read it fully rendered (headings, lists, code, bold). Filter the list by agent to see just one specialist's output.
                New Doc · edit · download · delete
                Create a document yourself, edit any doc in place, download it as Markdown, or remove it. Every action writes straight to disk.
                Where it lives
                Documents are plain Markdown files in the dashboard's own content/<agent>/ folders — nothing is hidden in a database, and nothing touches your Hermes home. source · content/ on disk
                Tab · Schedule

                Automations that run themselves

                Your recurring Hermes jobs — the routines that fire on a schedule (a morning briefing, a weekly brainstorm, and so on) — with the controls to manage them.

                The job list
                Each job shows its name, its schedule, the next time it will run, where it delivers, its last status, and the model it uses. source · Hermes cron/jobs.json (read-only)
                Run · Pause · Resume
                Trigger a job now, pause one so it stops firing, or resume a paused one — straight from the card. These actions go through your real Hermes.
                Top counters
                How many jobs exist, how many are active, and when the next run is due.
                Under the hood · Data & safety

                Real data, kept safe

                A few principles hold across the whole dashboard, so you can trust what you see.

                Read-only on your Hermes
                The dashboard reads your Hermes data (sessions, cron, gateway status) strictly read-only. It never writes to your Hermes home — it cannot corrupt or change your agents' state.
                Its own local files
                The only things it writes are its own: your task board, the activity log, your content documents, and timestamped backups. All live inside the dashboard's project folder. board.db · agent-logs.db · content/ · backups/
                Nothing fabricated
                Every number traces to a real source — your logs, your sessions, your server. There are no invented stats and no fake "savings" figures anywhere.
                Private by default
                The server binds to your machine only (127.0.0.1). You reach it remotely over your private Tailscale link — it is never exposed to the public internet.
                Everywhere · The top bar

                Controls you'll use on every tab

                Tab bar
                Switch between Overview, Agents, Chat, Tasks, Office, Content, Schedule, and this Docs page. The active tab is highlighted.
                Search
                Jump to agents, missions, or log entries from anywhere.
                Version badge
                The small v1.x next to the logo tells you which build you're looking at. It bumps every time the dashboard is meaningfully updated.
                Status & clock
                A live gateway-status pill and the current UTC time, so you always know the system is connected and what "now" means for the numbers.
                Audit

                Santé de la flotte

                —
                Total issues
                —
                Erreurs
                —
                Avertissements
                —

                Détail des issues

                Chargement...