Everything around Claude Code, handled.
AgentHub22 never wraps or proxies the model — it manages the sessions, accounts and telemetry around real, interactive Claude Code terminals.
Environments
Run every account side by side
An environment is a fully isolated Claude Code login — its own credentials, its own settings, its own tint color. Private and work are just two environments signed into different accounts. Each one carries default folders, a most-recently-used folder list, and a permissions preset: pre-approve specific tools, or flip on bypass mode for the environments you trust.
- Isolated login per environment (one-time /login, stored in your Keychain)
- Per-environment tool permissions: granular allow-lists or full bypass
- Tints, default folders and MRU folders per environment
Telemetry
See what every agent is doing — live
Every session streams its own telemetry whether or not its terminal is on screen: token totals split by input, output and cache, context high-water mark, a clickable tool-call timeline with full payloads, subagent activity, and the model's thinking stream. Sessions that need your input badge amber and can play a soft sound — you'll never discover a stuck permission prompt an hour late again.
- Full tool-call payloads, pretty-printed with copy buttons
- Subagent tracking: type, status, tokens, tool counts
- Waiting-state badges and optional notification sounds
Agent-to-agent
Your agents can talk to each other
AgentHub22 hosts a local MCP server that every session is automatically connected to. Agents can list the other live sessions and send them messages, delivered straight into the target's terminal as a visible, auditable prompt. The backend agent can tell the frontend agent the API contract changed — without you playing messenger.
- list_agents and send_to_agent tools, zero configuration
- Messages are visible in the receiving terminal — fully auditable
- Per-session accept toggle; loopback only, never cross-machine
Sessions
Close like tabs, resume like nothing happened
Closing a session archives it instead of losing it. Past sessions persist across app restarts with their title, environment, folder and final token totals — one click resumes the same conversation in the same environment and working directory, and the inspector rebuilds its stats by replaying the transcript.
- Past Sessions list persists across launches
- Resume targets the current conversation id, even after /clear or /resume
- AI-generated session titles from the transcript