A tech lead asked to give the team shared agents often assembles four systems: Hermes Agent (Nous Research) as the runtime and messaging gateway, Slack as the place people talk to it, an Obsidian vault as shared knowledge, and a memory store such as Mem0, Zep or Letta. Then come the scripts someone has to write: one to prune what piles up, one to report who is spending model credits.
Vocion is enterprise-grade AI you own: an open source platform with apps out
of the box, running in your AWS account or ours. One install covers most of
that list. Last verified 2026-10-08, against vocion-core v5.2.0.
This is not a case against the assembled stack. It is a good stack, and Hermes in particular is strong where Vocion is narrower.
Where the assembled stack wins
- Many more places to chat. Hermes' gateway reaches Telegram, Discord, Slack, WhatsApp, Signal, SMS, email, Microsoft Teams, Matrix and more from one background process (https://hermes-agent.nousresearch.com/docs/user-guide/messaging/). Vocion's chat surfaces today are its own apps and Slack.
- Its own command gate. Hermes checks each command against a list of dangerous patterns and asks the user to approve a match (https://hermes-agent.nousresearch.com/docs/user-guide/security).
- Pick each part. Swap the memory store or the notes tool without touching the rest. Letta, for one, offers stateful agents, shared memory, Slack, Telegram and Discord, and schedules in one open source project (https://docs.letta.com/overview). Open Notebook is MIT-licensed with a full REST API (https://github.com/lfnovo/open-notebook).
Row by row
| What the team asked for | The assembled stack | Vocion today |
|---|---|---|
| The same agents for everyone | Hermes profiles on one gateway | Agents authored as YAML in a git workspace and applied to one shared Postgres; teams group agents under a lead (docs/entities/agent.md, docs/entities/team.md) |
| Shared context | An Obsidian vault of Markdown | The same workspace: every workspace:apply records a workspace_version, and every tool call carries the workspace_sha that produced it (README.md) |
| Shared memory | Mem0, Zep or Letta | Learnings an agent reads; feedback becomes a candidate rule a person accepts or rejects (services/LearningCandidateService.ts) |
| Lives in Slack | Hermes' Slack gateway | Mention an agent in a channel or message it directly, and it replies in the thread. You bring your own Slack app; there are no Approve or Reject buttons in Slack, so proposals are approved in Vocion (docs/guides/slack.md, behind VOCION_SLACK_EVENTS=1) |
| Apps, not only chat | Not part of the stack | Workforce, Assistants, Software Factory and GTM |
| Your own model keys | Keys per profile | Per-workspace keys for OpenAI, Anthropic, Vertex, Azure OpenAI, AWS or a custom endpoint, encrypted with AES-256-GCM under a per-workspace key (libs/platforms/registry.ts, libs/crypto/credentialVault.ts) |
| Who is spending credits | A script you write | Per-agent budgets with soft caps that warn and hard caps that refuse a run before it starts (services/BudgetService.ts). Scoped per agent, not per person |
| Cleanup and recurring jobs | A cron job you write | Automations on a five-field UTC cron or an event, on durable schedules in the app's own Postgres (docs/entities/automation.md) |
| Where it runs | A VPS or machine you manage, plus each hosted service you choose | Your AWS account, Vocion Cloud, or self-hosted with Postgres 16 |
Which to pick
Take the assembled stack if your team lives in a messaging app Vocion does not reach, or you want to choose every part yourself. Take Vocion if you want one install with apps, shared agents, budgets and a record, running in an account you own. Nothing here rules out running both.
Related
- Compare approaches: closed SaaS AI, building on a framework, or Vocion.
- Slack surface, phase 1: what the Slack chat surface does.
- Learnings: how feedback becomes a rule.
- Teams: grouping agents under a lead.