Project review
A self-improving agent that actually remembers.
A serious OpenClaw peer when persistent memory and backend flexibility matter more than lowest-friction setup.
By ClawBlog Reviews Desk · Drafted with ClawBlog's research pipeline; edited and accountable to the named reviewer.
86
/100
ClawScore
Strong
82
/100
Users' Score
example avg 8.2/10
Review consensus
ClawBlog critics
A serious OpenClaw peer when persistent memory and backend flexibility matter more than lowest-friction setup.
86
/100
ClawScore
Strong
Review by ClawBlog Reviews Desk
Score profile
Readers
Approved reader ratings unlock at 5 ratings and stay moderated before publication.
82
/100
Users' Score
example avg 8.2/10 from 5
Hermes-Agent: I kept this in the loop for a month of normal feature work. The strongest part was how quickly it recovered from messy repo state without making the review process feel brittle.
Hermes-Agent: The setup path was smoother than expected and the defaults were sensible. I still had to tighten a few permissions for our workflow, but the day-to-day experience held up.
Example member reviews
Reader ratings for Hermes-Agent, sorted newest first by default.
Featured
Mara Chen
9/10 · 18 helpful
Hermes-Agent: I kept this in the loop for a month of normal feature work. The strongest part was how quickly it recovered from messy repo state without making the review process feel brittle.
Newest
Samira Okafor
8/10 · 8 helpful
Hermes-Agent: The review queue and audit trail mattered more than raw speed for us. It was easy to understand what happened, what changed, and when a human needed to step in.
Jun 14, 20268 review karma
Hermes-Agent: The review queue and audit trail mattered more than raw speed for us. It was easy to understand what happened, what changed, and when a human needed to step in.
8
/10
8
Helpful
Strong
Signal
Example reviews preview the voting surface, but they are not persisted and cannot receive votes.
Jun 13, 20269 review karma
Hermes-Agent: I liked the basic shape, especially once I stopped treating it like a magic button and gave it bounded jobs. The rough edges were mostly around longer-running context.
7
/10
9
Helpful
Mixed
Signal
Example reviews preview the voting surface, but they are not persisted and cannot receive votes.
Jun 12, 202612 review karma
Hermes-Agent: For repeatable agent tasks, the value showed up in the small things: clearer status, fewer surprise handoffs, and a useful paper trail when the output needed review.
9
/10
12
Helpful
Rave
Signal
Example reviews preview the voting surface, but they are not persisted and cannot receive votes.
Jun 11, 202614 review karma
Hermes-Agent: The setup path was smoother than expected and the defaults were sensible. I still had to tighten a few permissions for our workflow, but the day-to-day experience held up.
8
/10
14
Helpful
Strong
Signal
Example reviews preview the voting surface, but they are not persisted and cannot receive votes.
Jun 10, 202618 review karma
Hermes-Agent: I kept this in the loop for a month of normal feature work. The strongest part was how quickly it recovered from messy repo state without making the review process feel brittle.
9
/10
18
Helpful
Rave
Signal
Example reviews preview the voting surface, but they are not persisted and cannot receive votes.
The Criteria section is the detailed rubric behind the official ClawScore: each row carries its weight, rationale, and receipts.
Capability
Weight 1.6
Persistent memory, autonomous skill creation, MCP/tooling, gateway messaging, and subagents make Hermes-Agent a high-capability personal-agent harness.
Reliability
Weight 1.3
The receipts support stronger confidence than the first draft, while long-running recovery and memory hygiene still need a ClawLab pass.
Setup & DX
Weight 1.1
Desktop, terminal, portal setup, multiple backends, and Hostinger VPS packaging meaningfully reduce setup friction for a self-hosted agent.
Safety & Control
Weight 1.4
Command approval, authorization, container isolation, and MCP filtering are visible controls, but persistent memory remains a governance burden.
Cost Efficiency
Weight 1
Hermes-Agent is rated on cost efficiency from currently bound launch evidence. Unsupported details remain Analysis until receipts are attached.
Docs & Support
Weight 1
The official docs now cover install, memory, skills, messaging, MCP, security, architecture, and troubleshooting in enough depth to support publication.
Momentum
Weight 1.2
Nous stewardship, the canonical GitHub repository, and a packaged hosting path make momentum one of Hermes-Agent's strongest dimensions.
Hermes-Agent deserves a firmer launch score than the first draft gave it. The current primary receipts no longer describe a vague orchestration project; they describe a Nous Research agent with an explicit learning loop, persistent memory, broad tool surface, and a real deployment story. The official docs frame Hermes as an agent that builds memory and skills across sessions, runs through desktop or terminal flows, and lives across messaging platforms instead of staying trapped in one chat UI. That is a more complete product shape than the original review credited.
The strongest case is capability plus deployment range. Hermes supports a persistent memory system, autonomous skill creation, MCP/tool integrations, gateway messaging, scheduled automations, and isolated subagents. The runtime options are broader than the earlier draft assumed: local, Docker, SSH, Daytona, Singularity, and Modal appear in the docs, while Hostinger now offers a one-click VPS surface for teams that want self-hosting without hand-building the whole box. That puts Hermes at least on par with OpenClaw for teams who want a self-improving personal agent, and ahead in some memory-centric workflows.
The caveat is the same thing that makes Hermes interesting. Persistent memory is not only a feature; it is an accumulated state surface. An operator has to govern what the agent stores, what it learns from untrusted input, which tools it can call, and how self-improvement is reviewed before it becomes policy. The docs surface command approval, authorization, container isolation, and MCP filtering, which earns a stronger Safety & Control score than the first draft, but the review still does not treat "self-improving" as automatically safe.
The revised score therefore moves Hermes-Agent from watchlist to recommended-with-operator-discipline. It is not the easiest path for a casual user, and ClawLab still needs to test long-running reliability, memory cleanup, backend isolation, and recovery after tool failure. But the public receipts support a high score: official Nous stewardship, MIT licensing, active docs, substantial community momentum, a growing hosted setup path, and a coherent technical thesis around memory and skills. For builders willing to own the governance work, Hermes-Agent now looks like one of the strongest entries in the launch catalog.
Audience rating
Share one clear score and a short note. Approved reviews feed the Users' Score after moderation.
Checking session...
