Hermes Agent alternatives: pick by the workflow you are moving

Seven Hermes Agent alternatives hold up when checked against their own repositories on September 21, 2026: OpenClaw for a persistent multi-channel assistant, opencode and Pi for a terminal coding loop, OpenHands for a managed execution environment, NanoClaw for per-session container isolation, ZeroClaw for a minimal self-hosted footprint, and n8n when the job is a scheduled workflow rather than an agent. None of them is a drop-in replacement, because Hermes Agent is not retired and the candidates are not the same product category. Choose by the burden you are moving, then check what your memory, skills and channels do when they land — that part is where the surprises are, and Kunavo can only be the model endpoint underneath whichever one you pick.

Two corrections before the shortlist, because the roundups occupying this search result get them wrong. Hermes Agent is alive: the GitHub API returns archived=false for NousResearch/hermes-agent, created July 22, 2025, last pushed September 21, 2026, MIT licensed, with roughly 247,700 stars and a newest release named "Hermes Agent v0.21.3" under tag v2026.9.14, published September 14, 2026. And Hermes Agent does not run on the Hermes model weights: the repository's own Nous Portal guide carries the heading "Don't pick Hermes-4 for agent work" and calls those models chat and reasoning models, not tool-call-tuned. Separately, a "Hermes" result about bytecode and app startup is Meta's JavaScript engine for React Native, and one about parcels is the UK carrier that became Evri in March 2022.

Why people leave, and what each reason points at

Hermes Agent is free — its FAQ says plainly that the agent is free and open-source under MIT and that you pay only for the LLM API usage from your chosen provider — so nobody leaves over a licence fee. The reasons that actually come up are operational.

Two more verified projects sit just outside that list and have their own Kunavo comparisons: nanobot, MIT with release v0.3.5 on September 15, 2026, whose arbitrary custom provider name is OpenAI-compatible only; and Agent Zero, release v2.12 on September 9, 2026, which splits models into three roles — main chat, utility and embedding. That last role is worth naming out loud: Kunavo serves no embedding model and no speech-to-text model, so an embedding or transcription slot stays wherever you run it today rather than moving to a chat gateway.

The candidates, their licences, and who sells what around them

Repository status and release dates were read from the GitHub API on September 21, 2026. Two licence readings disagree with GitHub's badge and the LICENSE file wins.

Three notes that matter more than the table. The three things Nous sells around Hermes — the Portal subscription at $20, $100 and $200 per month for $22, $110 and $220 in monthly credits, Hermes Cloud at "Running $0.56/day" for a Medium instance and "Running $1.09/day" for a Large one, and the Tool Gateway — are separate products, never the price of the agent, and Hermes Cloud states that "Inference and tool usage are billed separately, on top of the instance price." The Tool Gateway is documented as a paid-subscription feature and publishes no per-call rate at all, so nothing here estimates one. And n8n's prices are quoted in euros on its own pricing page; they are not converted here. All checked September 21, 2026. Three stale identifiers are worth avoiding while you research: opencode-ai/opencode is a different, archived repository last pushed in September 2025, and All-Hands-AI/OpenHands and sst/opencode now redirect to OpenHands/OpenHands and anomalyco/opencode.

What migrates out of Hermes, asset by asset

Hermes ships a restore point, not an export. hermes backup creates a zip of configuration, skills, sessions and data while excluding the hermes-agent code itself, and hermes import restores that zip so archived files "overwrite existing files in your Hermes home" — both per the CLI command reference, read September 21, 2026. There is no top-level hermes export, and hermes migrate is a different thing entirely: it rewrites config.yaml to replace retired models and deprecated settings. So the only documented route out is an importer on the other side, and exactly one project was checked exhaustively for one.

The mechanics, from OpenClaw's Hermes migration page and its migrate CLI reference, both fetched September 21, 2026: preview with openclaw migrate hermes --dry-run, apply with openclaw migrate apply hermes --yes, and read the plan before adding --overwrite or --include-secrets. There is a hard precondition — "Imports require a fresh OpenClaw setup. If you already have local OpenClaw state, reset config, credentials, sessions, and the workspace first" — and a backup is created and verified before applying. No rollback or undo command is documented on either page. Your rollback is therefore procedural: run hermes backup first, leave Hermes installed and paired while you trial the replacement, and move one workflow at a time. For the opposite direction, Kunavo already covers migrating OpenClaw to Hermes, and Hermes vs OpenClaw compares the two head to head.

For NanoClaw, ZeroClaw, nanobot, Agent Zero, OpenHands and n8n, no Hermes importer was found — but only OpenClaw's documentation was searched exhaustively, so read that as "no import path found at that scope", not as a proven absence.

The API bill follows you, and two clients report it as zero

Whichever project you land on, the recurring cost is model tokens. Before comparing rates, check the protocol ceiling: Hermes accepts exactly three transport values on a provider entry — chat_completions, anthropic_messages and codex_responses — with URL-based auto-detection documented only as "a fallback when the field is left blank", per the providers page read September 21, 2026. Pi accepts four, adding google-generative-ai and splitting OpenAI into completions and responses. An endpoint that speaks none of a client's protocols is not configurable as its provider at any price.

Then the trap that makes cost comparisons lie. Pi's model configuration table gives the per-model cost field a default of "all zeros", so a hand-added custom provider reports $0 spend until you type the rates in; the same table defaults contextWindow to 128000 and maxTokens to 16384. Kunavo's own earlier notes record OpenClaw booking a custom model at zero cost for the same reason — that one is carried from pages dated September 18 and 21, 2026 rather than re-read here. In both cases the client's spend readout is its own arithmetic, not your provider's ledger.

Here is what the token side looks like at live Kunavo catalog rates. These are illustrative token arithmetic, not measured task costs and not a bill ceiling. Assume one agent session sends 200,000 uncached input tokens and receives 12,000 output tokens, and assume 80 such sessions in a month. Cache charges, tool calls and retries are excluded, and no client on this page was benchmarked to produce these token counts.

Read that against the subscriptions rather than instead of them: the cheapest listed rate and the lowest cost to finish a task are different claims, and a model that needs three attempts can cost more than one that needs one. Kunavo's catalog amount is a billing floor rather than a cap — when the upstream reports its charge, the bill is the greater of catalog cost and upstream cost times the applicable markup. The minimum entry is $10 in prepaid credit, a funding minimum rather than a task fee or a subscription, and there is no plan gate on any Kunavo feature; see billing details. Per-key monthly spend limits and per-key IP allowlists are configurable at your keys page, which is the practical guardrail when an unfamiliar agent is looping unattended.

A safe order of operations

Run hermes backup, keep Hermes installed and its channels paired, and give the replacement one bounded task before you move anything else. Point it at https://api.kunavo.com/v1 with the wire protocol written by hand rather than left to auto-detection, run the task, then read the charge your account actually recorded for it — not the client's own spend line, which may be reporting zero. Kunavo has not runtime-tested Hermes Agent or any of the clients on this page against its endpoint; a published setup guide is a configuration reference, not a compatibility test.

Start at the opencode integration guide if the terminal coding route is the one you picked, and create a Kunavo account when you are ready to fund a key. If you have not settled the client question yet, Hermes custom API setup covers staying put with a different endpoint, Hermes coding agents covers the Codex runtime trade-offs inside Hermes, and OpenClaw alternatives runs the same decision from the other side.

FAQ

Repository metadata for Hermes Agent and nine alternatives was read from the GitHub API on September 21, 2026, and every vendor page linked above was fetched the same day: the OpenClaw Hermes-migration, migrate CLI and skills pages; the Hermes providers, CLI-command, memory and skills pages; the Nous Portal, Hermes Cloud and Tool Gateway pages; opencode's Go and Zen pages; OpenHands' and n8n's pricing pages; and the Pi, ZeroClaw, NanoClaw, nanobot and Agent Zero docs. Star counts move daily and are reported as read that day. The OpenClaw zero-cost custom-model note is carried from Kunavo's earlier pages. Kunavo token rates read from the live catalog; every dollar example is illustrative token arithmetic.