Overview
Agent Hub is the central home for agent work in the Allternit platform. It consolidates the agent builder, running session history, performance analytics, and per-agent workspace files into one tabbed surface so you can move from prototyping to production without switching contexts. The hub is registered as a singleton view in the platform shell. Only one instance stays open at a time, and it communicates with the shared agent store, mode-session stores, and the agent metrics service.How to access it
Open Agent Hub from the left rail by selecting Agent | Bot Hub, or navigate directly:initialTab prop so external callers can land directly on Studio, Sessions, Analytics, or Workspace.
Hub tabs
Features
Agent Studio
Visually build an agent: name, system prompt, model, temperature, max tokens, and tool bindings. Run a prototype conversation and save when ready.
Session Manager
Search and filter agent sessions across all mode surfaces, then resume exactly where you left off.
Performance Analytics
Track total runs, average latency, tokens consumed, estimated cost, and per-agent success rates over 1H, 24H, 7D, or 30D windows.
Agent Workspace
Browse and edit an agent’s bootstrapped files, including
agent.config.json and .allternit/* configuration and identity artifacts.Singleton View
Registered as a single-instance shell view so the hub is always one click away and never litters the tab bar.
Rails-Integrated State
Runs, tasks, checkpoints, commits, queue items, and mail threads sync through the agent service and Rails event stream.
Key concepts
Mode sessions
A mode session is a conversation or working session started with an agent inside one of the platform’s mode surfaces. The Sessions tab reads agent-tagged sessions from the chat, code, cowork, and design session stores and surfaces them in a single searchable grid. Clicking a session restores it in its original surface.Agent registry and local fallback
The hub loads agents from the platform’s agent registry through the Agents API. If the registry is unreachable, it transparently falls back to the local agent registry so the UI remains usable offline or during a backend rollout.Character layer
Every agent carries a character layer that defines identity, role card, voice, progression, and avatar. The store validates this layer on create and update, and it checks hard bans before a run starts.Metrics service
Analytics are fetched from/api/v1/agents/metrics with a days parameter driven by the selected time range. The summary aggregates total runs, average latency, tokens consumed, total cost, and success rate per agent.
Workspace files
The Workspace tab mounts the sameWorkspaceTab used in the Agent Dashboard and Settings → Agents. After selecting an agent, you can inspect its workspace directory (typically agents/{agentId}) and edit configuration, memory, and governance files.
Endpoints used
Related docs
- Platform — the main web shell that hosts Agent Hub
- Agent Studio — deep dive into the agent builder tab
- Bots — surface for bot discovery and management
- Swarm Dashboard — monitor multi-agent swarms
- Agents API — backend reference for agent CRUD and runs
- Gizzi Code CLI — local agent tooling and runtime