Summary

Allow a single Telegram bot to route messages from different forum topics (threads) to different Hermes profiles, so each topic can be handled by a specialized agent with its own model, skills, memory, and system prompt.

Current Behavior

• Each profile runs its own gateway with its own bot token → 1 bot = 1 profile
• Within a profile, build_session_key() creates separate sessions per thread_id, but all threads are still handled by the same agent (same model, same skills, same system prompt)
• To have different agent behavior per topic, users must run N separate bots (N Telegram bot tokens, N gateway processes), which is impractical for large setups

Proposed Behavior

Add config-based routing so that a single gateway can dispatch incoming messages to different profiles based on chat_id + thread_id:

# In the gateway profile's config.yaml
telegram:
  topic_profiles:
    - match:
        chat_id: "-1001234567890"
        thread_id: "1939"
      profile: "coding-assistant"
    - match:
        chat_id: "-1001234567890"
        thread_id: "42"
      profile: "research-agent"
    - match:
        chat_id: "-1001234567890"
        thread_id: "99"
      profile: "polymarket-trader"

When a message arrives, the gateway:

1. Checks topic_profiles for a matching chat_id + thread_id
2. If matched, spawns/forwards to the target profile's AIAgent (different model, skills, SOUL.md, memory)
3. If no match, falls back to the current behavior (handle with own profile)

Why This Matters

• Single bot, multiple specialists: One Telegram bot serving a community group, where each forum topic has a purpose-built agent (coding, research, trading, CS)
• Resource efficient: No need for 10 separate bot tokens and 10 gateway processes
• Better UX: Users stay in one group, interact naturally via topics — no need to DM different bots

Implementation Notes

• The routing layer would sit in gateway/platforms/telegram.py between message receipt and session dispatch
• The target profile's AIAgent could be spawned in-process (load profile config + create agent instance) or forwarded via IPC
• Session keys would need to incorporate the target profile to avoid collision
• This is conceptually similar to dm_topics (which routes DMs to forum topics), but at the profile/agent level

Alternatives Considered

1. Multiple bots — current workaround, but requires N tokens and N processes
2. Dynamic skill/model switching per topic — possible via extra.topic_config in gateway, but doesn't isolate memory or system prompts
3. External proxy/router — adds complexity and latency

Related

• gateway/session.py — build_session_key() already handles thread_id isolation
• gateway/platforms/telegram.py — _dm_topics config pattern as precedent
• Multi-agent profiles discussion: https://x.com/teknium/status/2044100748963918108