Hermes Agent from Nous Research is a general agent that remembers you across sessions, writes its own skills, runs scheduled jobs and answers in Telegram, Slack and about 20 other chat apps. OpenCode from Anomaly is an open-source coding agent for the terminal, desktop and web. Both are free and MIT licensed; you pay for the model. They also connect: Hermes can run the OpenCode CLI as a coding worker, and it can use an OpenCode Go plan as its model.
Checked against Hermes 0.23.0 (November 5, 2026) and OpenCode 2.2.6 (November 6, 2026), using each project's own docs and source code.
| Capability | Hermes AgentNous Research · MIT | OpenCodeAnomaly · MIT |
|---|---|---|
| What it is | A general agent with memory, skills and scheduled jobs | A coding agent |
| Where you use it | Desktop app (macOS, Windows), terminal, web dashboard and chat apps | Terminal, desktop app (macOS, Windows, Linux), web, and an iPhone beta |
| Working while you are away | Cron jobs that send results to a chat, a file or another platform | Opt-in scheduled prompts since 2.2.1, still experimental and limited to one session |
| Reaching you in chat apps | Built-in gateway for Telegram, Discord, Slack, WhatsApp, Signal, email and about 15 more | None of its own |
| Memory across sessions | MEMORY.md and USER.md notes, plus search over past sessions | AGENTS.md rules and session history; no memory store |
| Skills | Writes its own skills after hard tasks and improves them as it uses them | Loads skills from .opencode, .claude and .agents folders; does not write them |
| Coding tools | File edits checked by language servers; up to 10 subagents at once | A Plan agent that cannot edit files, a Build agent that can, subagents, /undo and git worktrees |
| Models | 43 listed providers, any OpenAI-compatible endpoint, local models and Nous Portal | 200+ providers from models.dev, and local models through Ollama |
| Plans you can sign in with | ChatGPT, OpenCode Go; Claude only through Max extra-usage credits | ChatGPT Plus or Pro, GitHub Copilot, and a Claude login by running your installed Claude Code |
| Approvals | Checks commands against a list of dangerous patterns; YOLO mode skips the check | Allow, ask and deny rules; by default everything is allowed, and reading .env files asks first |
| Windows | Runs natively on Windows 10 and 11, no WSL needed | Native Windows command-line and desktop builds |
Neither approval system is a sandbox, and both projects say so: Hermes calls the operating system the only real security boundary, and OpenCode suggests a container or a VM if you need isolation. Many comparisons online still describe OpenCode 1, where language servers, the /opencode GitHub agent and share links lived; OpenCode 2 dropped or moved them.
This comparison was last updated on November 8, 2026; check each vendor's site for current terms.
Most people searching this end up running both. The connections all go one way: Hermes uses OpenCode, and OpenCode has nothing that runs Hermes.
An official Hermes skill runs opencode run for one-off tasks, or drives the OpenCode interface in the background for longer ones. Since Hermes 0.22.7 it is an optional install rather than a default one. Its setup line installs OpenCode 1 (the opencode-ai package); the core commands it uses also exist in OpenCode 2.
hermes skills install official/autonomous-ai-agents/opencodeHermes has built-in OpenCode Go and OpenCode Zen providers, and OpenCode's own docs list Hermes as a validated client. Go costs $10 a month (Go Plus $40) and now includes some closed models alongside open ones. It needs Hermes 0.21.6 or later. OpenCode's free models only work inside OpenCode and return HTTP 403 from Hermes.
hermes model # choose "OpenCode Go", paste your Console key
hermes chat --provider opencode-go --model kimi-k3OpenCode reads ~/.agents/skills on its own. Hermes reads it only after you add it to skills.external_dirs in ~/.hermes/config.yaml, and it may then edit those skills where they are.
# ~/.hermes/config.yaml
skills:
external_dirs:
- ~/.agents/skillsOrkas runs Hermes and OpenCode as two local agents in the same chat, next to Claude Code and Codex. The next table lists what Orkas can and cannot hand each of them.
These rows come from Orkas's adapter code for each CLI, which is the same in Orkas 2.0 and the next release. They describe what a host can do with each agent, not how well either one codes.
| In Orkas | Hermes Agent | OpenCode |
|---|---|---|
| How Orkas starts it | hermes acp, over the Agent Client Protocol | opencode run, one process per turn |
| Project folder | Works from the Orkas workspace | Bound to the project folder you choose |
| The next turn | A fresh session; Orkas resends the visible conversation | Continues the same session |
| Permission level | Use CLI default, Ask for permission (you approve in Orkas) or Full access | Full access only; Orkas asks you to confirm it when you connect OpenCode |
| Orkas skills and the services you connected | Available in every run | Available in every run |
| Sending a message mid-run | Waits until the turn ends | Waits until the turn ends |
Hermes's own ACP server can now resume sessions; Orkas does not use that yet, so it resends the history. OpenCode's one-shot run has no way to answer an approval prompt, which is why Orkas runs it with full access. Orkas checks for Hermes 0.9.0 or later and OpenCode 1.18.0 or later, finds both on your PATH, and does not install either.
Orkas is a free, MIT-licensed desktop app for macOS and Windows. It runs the Hermes and OpenCode you already have, plus Claude Code, Codex and OpenClaw, as agents in one chat.
Install Hermes and OpenCode first and sign each one in to a model. Orkas finds them on your PATH.
Open Connect agents and add both. When you add OpenCode, Orkas asks you to confirm full access.
Set Hermes's approvals in AI Team > Hermes > Runtime settings > Permission level, then give each agent its part of the task in the same chat.
What Orkas does not do here: it does not install either CLI, resume a Hermes session or show OpenCode's own approval prompts. Hermes's cron jobs and chat gateway stay in Hermes.
It can code: it edits files, checks its edits with language servers and runs up to 10 subagents at once. But it is built as a general agent, and for hands-on work in one repository most people pair it with a coding CLI. Hermes ships skills for Codex and Claude Code by default, and an optional one for OpenCode.
Yes, in two ways. Install the official skill with hermes skills install official/autonomous-ai-agents/opencode and Hermes runs opencode run or drives the OpenCode interface for coding tasks. Or choose OpenCode Go or OpenCode Zen as Hermes's model provider with an OpenCode Console API key. Since Hermes 0.22.7 the skill is no longer installed by default.
Yes. Run hermes model and choose OpenCode Go, then paste an OpenCode Console API key; Hermes saves it in the ~/.hermes/.env file. Go needs Hermes 0.21.6 or later, which sends the session header OpenCode requires. Go costs $10 a month, or $40 for Go Plus. OpenCode's free models only work inside OpenCode and return HTTP 403 from Hermes.
Both are free and MIT licensed, so the cost is the model. Either one can use a ChatGPT plan. Hermes adds Nous Portal: free pay-as-you-go, or $20, $100 or $200 a month, with a hosted Hermes Cloud agent on the paid tiers. OpenCode adds OpenCode Go and pay-as-you-go Console credits. Hermes cannot use a Claude Pro or Max allowance, only Max extra-usage credits.
Yes. Orkas runs both as local agents in one desktop window, next to Claude Code, Codex and OpenClaw. OpenCode works in the project folder you choose and continues its own session. Hermes starts a fresh session each turn, and Orkas resends the conversation so it keeps the context.
Orkas is free and MIT licensed. It drives Claude Code, Codex, OpenCode, OpenClaw and Hermes as local sessions on your own machine.