Hermes starts moving from Kanban coordination to one gateway running many named agents
The most product-shaped Hermes update in this window is PR #25008: a single-gateway multi-agent MVP. It adds `agent_id` to session sources and session rows, introduces an `AgentProfile` ContextVar so model, SOUL.md, memory, skills, and session paths can switch per named agent, routes platform events by chat / thread / user / guild with a first-match-wins matcher plus a plugin hook, and wires the same profile switching through adapters, cron jobs, delivery targets, hooks, and a new `hermes agent` CLI. The surrounding fixes show why this architecture needs careful runtime hygiene: #25344 filters Honcho memory deriver noise such as “Nothing to save” before it reaches the main prompt, #25346 splits concatenated streamed tool-call argument blobs like `{...}{...}` into separate calls, #25341 cuts the `hermes tools` all-platforms menu from about 14 seconds to under 1.5 seconds while avoiding repeated Nous refresh-token burns, #25334 bypasses system proxies for localhost auxiliary clients, #25342 silences background-review memory-provider teardown output, and #22648 keeps an Ollama Cloud web backend moving forward.
Kanban gave Hermes a shared task board; this PR starts giving the gateway separate named workers with their own identity and state. That is a bigger operational shift than another UI feature, because routing mistakes now mean the wrong agent, memory, skill set, or cron job can answer. The nearby fixes are small individually, but together they are the plumbing that keeps multi-agent routing from becoming context contamination.
- PR #25008 adds single-gateway multi-agent routing with `SessionSource.agent_id`, `SessionEntry.agent_id`, `AgentProfile`, ContextVar-based path getters, declarative route matching, a `select_agent` hook, cron and delivery `agent_id`, and a new `hermes agent` CLI
- The PR reports 62 new tests for routing, ContextVar isolation, and session keys, while preserving the default `agent:main:...` key format for single-agent installs
- PR #25344 filters Honcho deriver housekeeping text before it is injected as memory context and labels valid contextual analysis
- PR #25346 repairs streamed tool-call assembly when concatenated JSON argument blobs represent multiple calls
- PR #25341 reports `hermes tools` all-platform rendering falling from roughly 14 seconds to about 1.25 seconds cold / 17 ms warm and refresh-token use dropping from 31 to 1 per render
- PRs #25334 and #25342 address localhost proxy inheritance for auxiliary clients and background-review teardown output leaking to the terminal
- The multi-agent MVP is an open PR, so installed builds may not include it and details may change before merge
- First-match route semantics are simple, but teams will need tests for overlapping chat, thread, user, and guild rules
- Profile isolation increases the blast radius of subtle context bugs: a wrong route can select the wrong memory, skills, cron ownership, or delivery identity