Timon: my review of the workstation that runs your AI agents
Timon shows 167 stars on GitHub and sits at version 0.1.7. The public repository was created on November 6, 2026, the day before I downloaded the installer. That is the first thing to say, before any talk of features.
Because Timon does not pitch itself as one more agent. The site calls it an “agentic workstation”, a desk that runs the agents you already use: Claude Code, Codex, Grok, GitHub Copilot, opencode, Hermes. So my review is less about a model than about where the work happens.
The Lab verdict
IN TEST7/10
Timon brings Claude Code, Codex, Grok, Copilot, opencode and Hermes into one workstation, with a git worktree per task and a cross review by another model. The architecture is the right answer to terminal chaos, but the product is one day old: version 0.1.7, repository created the day before my test.
- Real price : Free for 1 developer on 1 machine, all agents included, no support. Pro and Cloud are listed as “pricing coming soon”. Timon never resells tokens: you bring your own subscriptions.
- Best for : A solo developer already paying for several AI subscriptions and losing the thread between terminals, branches and conversations.
- Skip it if : You want a stable tool for a team, or a single turnkey agent with zero configuration.
Timon at a glance
Timon is a desktop app (Electron) backed by a daemon running as a system service. It gathers several command line agents into a single space, with a pipeline that runs from the job description to the merge, including a review pass by another model.
- Real price: free on the Free plan for one developer on one machine, all agents included, no support. Pro and Cloud are listed as “pricing coming soon”.
- Who it is for: a solo developer already paying for several AI subscriptions and losing time juggling terminals.
- Walk away if: you want a stable tool for a team, or a single turnkey agent.
The IN TEST badge is not a stylistic precaution: it follows directly from my test, which covered the package, the CLI and the documentation, without the graphical interface.
Timon is not another agent
Timon does not sell a model. It sells the desk the models work on.
You plug in your existing subscriptions, your API keys or a custom endpoint. Claude Code, Codex, Grok, GitHub Copilot, opencode and Hermes are listed, with “and more” next to them.
Two details change the way it feels: switching models mid-thread without losing context, and fallback between providers. When one provider hits its limit, the conversation continues on the next configured one, same thread, full context. The house slogan, “Never out of quota”, translates plainly: no more pasting your last message into another terminal.
The code, review, merge pipeline: the real differentiator
One block stopped me: the pipeline.
You describe the work. Timon proposes a plan, which you approve. The plan breaks into tasks on a Kanban. Each task is coded, reviewed and merged in its own git worktree. One model writes while another reads, what the editor calls cross-model review.
The worktree per task is the detail that matters. Two parallel tasks do not step on each other, and a review that goes wrong does not leave your folder in a strange state. That is plumbing, not magic, and it is exactly what is missing when three agents run side by side.
Cross review has its limits. Two models often share the same blind spots. It catches obvious regressions, not debatable architecture choices. Timon prepares your review, it does not replace it.
| Criterion | Timon | A CLI agent alone |
|---|---|---|
| Agents | Several, in one space | One per terminal |
| Model switching | Mid-thread | New thread, context to paste back |
| Isolation | One worktree per task | Managed by hand |
| Review | One model reads another | Arranged yourself |
What runs while you sleep
The Schedule queue sends tasks at 2 a.m. and spreads them until you wake up: closing the window does not stop the work, the daemon runs as a system service. Triggers cover cron, GitHub and GitLab webhooks, and internal events. Workflows offer a visual canvas with a trace you can replay node by node, and Team puts an orchestrator above coders, reviewers and mergers.
That is the scenario many people are after: start the machine in the evening and find branches ready in the morning. On paper only.
My real test: the CLI works, the interface is still unknown
I downloaded the Linux AppImage, 208 MB, and extracted it. The contents are consistent: resources/app at 0.1.7 under AGPL-3.0-only, resources/orchestrator for the engine, a cli/timon binary, the usual Electron binaries.
Then timon doctor inspects the machine: it detects node, git, docker and hermes-acp, then lists the missing backends, claude, codex, grok, copilot, opencode and gh. It reports the disabled modules (systemd, gui, containers) and closes with “Nothing missing at the essential level (node, git, an LLM backend)”.
The CLI also exposes more than the interface: timon spawn, send, steer, kill, the tasks, the skills, the MCP servers, the workflows and the traces. The limit is clear. The daemon refuses to start outside its Electron wrapper: the native module better-sqlite3 is compiled for the Electron ABI (NODE_MODULE_VERSION 127), not for the system Node. The consequence: no engine on its own, headless, without the app.
My environment had no graphical session, so I did not see the interface running. My review of Timon covers the package, the CLI, the documentation and the pricing.
The price: free, with a precise definition
The Free plan covers one developer on one machine. It includes all the agents, your subscriptions or keys, the Kanban with git, triggers, routines, workflows, the pipeline with one worktree per task, the multichat grid (2, 4, 6 or 9 conversations), the built-in browser, local memory and the mobile interface. One user, one machine, no support.
Pro and Cloud are listed as “pricing coming soon”, contact through Discord. Pro adds premium plugins, multi-user accounts with roles, white label and premium support. Cloud is Pro on your own server, the team connected from any browser, agents running around the clock. The editor says it plainly: “Plans and their contents may still change.”
The economic rule is stated without hedging: “Bring your own AI subscriptions: Timon never resells model usage.” Timon does not resell tokens. I already broke down the real bill of an agent running 24/7 if you want actual numbers before wiring one more agent. Your cost is your current subscriptions, plus the price of Timon once it is published.
Importing from Hermes, the underrated strength
timon agent import –hermes [PROFILE…] I explained why I quit n8n for my own AI agents, which is the reasoning behind tools like this one. reads the Hermes profiles on the machine, in ~/.hermes. The command pulls the prompt, the rules, the skills, the MCP servers, the routines and the image. Secrets are imported by name only.
If you already have agents configured, you do not start from scratch. Timon runs Hermes as one of its agents, next to Claude Code, Codex, Grok and opencode. It is the most concrete demonstration of its promise: it does not try to kill your agents, it hosts them.
Frequently asked questions
Is Timon really free?
Yes, with a restrictive definition. The Free plan covers one developer on one machine and includes all the agents, the Kanban, triggers, workflows, the pipeline with one worktree per task and the mobile interface. No support. Pro and Cloud are announced as “pricing coming soon”, with no public figure to date.
Does Timon replace Claude Code or Codex?
No. Timon is a workstation that runs those agents. You keep your subscriptions and your command line habits, but you drive them from one space, with a pipeline and task tracking on top. The tool organizes the work, it does not replace the engine that writes the code.
Do I need new AI subscriptions?
No. Timon works with your subscriptions, your API keys or a custom endpoint. The stated rule is explicit: Timon never resells tokens. Your variable cost stays what you already pay your model providers, and the tool takes nothing on the way through.
Can Timon run without the desktop app?
No, not in the version I tested. The daemon refuses to start outside the Electron wrapper, because the native module better-sqlite3 is compiled for the Electron ABI and not for the system Node. The engine stays tied to the desktop installation, even though it keeps working when the window is closed.
Can I import my existing Hermes agents?
Yes, with timon agent import –hermes. The command reads the profiles in ~/.hermes and retrieves the prompt, the identity documents, the rules, the skills, the MCP servers, the routines and the image. Secrets are imported by name only. Timon then runs Hermes as one of its agents, alongside Claude Code, Codex or Grok.
Is Timon reliable for daily use?
Not proven yet. The public version is 0.1.7, the repository was created on November 6, 2026, and it counted 167 stars, 17 forks and 10 open issues as of November 7. That is a fast start, not a track record. Hence the IN TEST badge: the product is promising, its durability remains to be proven.
The last word
Timon attacks a real problem from the right end. Most coding agent tools sell a model or a chat window. Timon sells the workshop: the worktrees, the Kanban, the cross review, the daemon that works at night, the import of what you already have.
Free is honest, the bring your own subscriptions rule is honest too, and the Hermes import shows the editor knows who it is talking to. But this is a repository created the day before, a 0.1.7 version and an interface I could not see running. A 7/10 in IN TEST is exactly that: an architecture worth a try, not a firm recommendation.
If you already have two agents installed and a habit of losing context between terminals, Free can be tried with no financial risk: install it, plug in one agent, let a real task run. If you are looking for a tool to put in every hand on a team, wait until Pro and Cloud have a price. And if you want a single agent, one command, zero configuration, you are not the audience.
Leave a Reply