Feature
One Hermes gateway, separate channel checks
Quick answer
A Hermes gateway connects configured messaging platforms to the agent runtime. Shared profile settings and memory do not mean every channel is the same live conversation. Verify the sender, session and exact destination on each platform; a connected adapter alone does not prove a completed reply.
Key Points
- ✓One gateway can host several configured platform adapters
- ✓Each channel keeps its own authorization and delivery requirements
- ✓Conversation/session identity is distinct from persistent profile memory
- ✓Telegram defaults to polling; webhook mode is an explicit alternative
- ✓Status and reactions are checkpoints, not end-to-end proof
- ✓Recovered replies can be labeled as possible duplicates after ambiguous sends
How It Works
- 1Configure the intended platforms in the owning profile without publishing credentials.
- 2Start the owning gateway and verify provider access; avoid competing processes for the same bot token.
- 3Send one harmless unique marker from an authorized account on each channel you intend to use.
- 4Verify the reply in that exact DM, group or thread and inspect the matching session before handing work across interfaces.
Real-World Use Cases
Mobile research with an explicit handoff
Start a read-only request in Telegram. Before continuing elsewhere, record the task, result location and session identity; use the supported session controls and access rules rather than assuming the conversation moved automatically.
Trusted team notifications
Deliver an approved report to a specific group or thread. Grant only the intended users or group access, and test that destination independently from a private DM.
Gateway health triage
Use the private dashboard for profile, provider and log inspection. Then test inbound receipt, authorization, model completion and final sending with a harmless marker.
Recovery without repeating external work
If a response repeats, inspect competing gateway ownership and recovered-reply labels. A duplicated message is not proof the underlying job should run again.
Under the Hood
Telegram uses polling by default and does not require a public inbound port in that mode. Webhook mode needs a public HTTPS endpoint and TELEGRAM_WEBHOOK_SECRET. BotFather privacy controls delivery of group messages; Hermes sender authorization and mention rules are separate checks. Messages sent while the host is off also depend on the configured cold-boot pending-update policy.
Shared profile memory and skills are not a promise of automatic live conversation synchronization. Current gateway documentation scopes /sessions to the current chat, with cross-origin listing restricted to administrators. Telegram DM topics have their own sessions. Before switching interfaces, identify the session, working context and expected output, and respect the access controls of the destination. Separate profiles isolate configuration and credentials, not the operating-system account by themselves.
The gateway delivery ledger stores final responses around platform sends. After a crash, it can redeliver a stored reply without rerunning the whole agent turn. An ambiguous mid-send response is labeled as a possible duplicate. This is separate from a message Telegram held while the gateway was offline and from in-memory suppression of repeated inbound updates. Reconcile external actions before retrying them.
Self-hosted WebUI includes PTY/WebSocket Chat as well as configuration and monitoring. It does not remove responsibility for runtime uptime, credentials, networking and recovery. Managed hosting changes who operates that infrastructure; verify the current channel support and plan terms before choosing.
Frequently asked questions
Does switching platforms resume the same live conversation?
Do not assume that. Profile memory can be shared while conversation sessions remain distinct. Current /sessions listing is chat-scoped; cross-origin listing is restricted to administrators. Verify session identity and use the supported handoff or resume controls.
Does Telegram require webhook mode?
No. Telegram defaults to polling, which needs no public inbound port. Webhook mode is an explicit alternative requiring HTTPS and TELEGRAM_WEBHOOK_SECRET.
What proves a channel is healthy?
An authorized sender submits a harmless request and receives the expected answer in the exact destination. Adapter status or a processing reaction alone cannot prove model completion and final sending.
Why might a reply arrive twice?
Check competing gateway ownership and the recovered-reply label. An ambiguous interrupted send can be retried by the delivery ledger. Inspect the external result before repeating the task.
Do all channel users get the same tool access automatically?
Do not assume that. Review platform and profile configuration, tool availability and sender authorization. For different trust levels, use deliberate profile and runtime isolation rather than relying on separate chat topics.
Is managed hosting required for WebUI Chat?
No. Self-hosted WebUI supports PTY/WebSocket Chat. Managed hosting is an operational choice about uptime, updates and recovery, not a prerequisite for browser conversation.
Next setup steps
Telegram setup and offline-message handling
Configure access, verify a reply and choose what happens during downtime.
Telegram connection checklist
Connect one bot before adding groups or topics.
Gateway troubleshooting
Separate transport, authorization, model and send failures.
Private WebUI and browser Chat
Inspect the runtime without confusing browser and messaging sessions.
Managed hosting comparison
Decide who should operate the runtime and check current terms.