KinetAios vs Claude Code: an honest look at the open-source alternative

People searching for a "Claude Code alternative" usually want one of two things: a different workflow (GUI, history, multiple agents), or a different engine under the hood. This post is written by the author of KinetAios, so instead of marketing, here is exactly what it does, what it doesn't, and who it's for.

What Claude Code does well

Claude Code is the strongest terminal coding agent available. Its agent loop, tool use, skills and hooks run inside Anthropic's own CLI, tightly integrated with the model. If you live in the terminal, work on one project at a time, and only use one engine — Claude Code alone is enough. Genuinely. Close this tab.

Where a terminal hits friction

After months of daily use, the friction isn't the agent — it's the cockpit:

• History is scattered. Sessions land as per-project JSONL under ~/.claude/projects/…. There is no cross-project, cross-day search over what you and the agent actually did.
• One REPL at a time. Running three agents in parallel means tmux panes, and switching contexts mid-thought.
• Nothing accumulates. Session N doesn't know what you decided in session N-3 unless you paste it back in yourself.
• One engine only. Trying Codex means learning a second REPL with different flags, different resume semantics, and a different history store.

What KinetAios is

KinetAios is a local-first desktop dashboard for coding agents on macOS and Windows. Open source (MIT), no account, no relay. The honest framing: it does not replace Claude Code's agent brain — it wraps it in a cockpit, and lets other engines share the same cockpit.

How it actually runs Claude Code

No reimplementation, no proxy. KinetAios spawns the real CLI and parses its structured stream:

claude -p --output-format stream-json --verbose --resume <sessionId>

Every NDJSON line — tokens, tool calls, costs, status — is folded into a unified event model. Resume uses Claude Code's own session id, so you can drop back to the terminal and continue the same session. Your stored preferences ride along via --append-system-prompt. The agent loop is still Anthropic's; the dashboard adds the seats around it.

Codex gets the same treatment (codex exec --json, same resume pattern). And the built-in "Direct" engine is a compact ReAct loop for when you want a second opinion from a different provider without leaving the window.

What it deliberately doesn't do

• It doesn't fork or proxy Claude Code's internals. Skills, hooks and sub-agents still live inside the CLI and keep working — KinetAios doesn't intercept them.
• No local LLM endpoint yet. There is no Ollama / OpenAI-compatible base-url setting today. If local models are your hard requirement, this isn't the tool for that yet — the roadmap has it, but shipping a half-working local endpoint would be worse than not shipping it.
• It's not a team product. No cloud sync, no seats, no relay server. Everything stays in your local SQLite file.

Who should try it

• You run Claude Code daily and want searchable history across projects instead of JSONL archaeology.
• You juggle Claude Code and Codex and want one cockpit instead of two REPLs.
• You want your agent's decisions to accumulate into long-term memory automatically.
• You're on Windows and prefer a native desktop app over a terminal window.

If that trade sounds right — the agent brain you already trust, in a cockpit that remembers — that's the whole pitch. If it doesn't, the field is wider than one app: cloud dashboards, terminal multiplexers, cross-platform workbenches. I keep the 2026 Claude Code GUI comparison updated separately, with the same disclosure rules.