Hermes IDE
An IDE that integrates multiple AI coding agents, supports switching between Claude Code and Meta models, and can run tasks in separate git worktrees.
Timeline (last 90 days)
Hermes IDE version 2.2.9 fixed a related problem: cost and token totals kept growing while a session was open because per-session usage was counted multiple times; totals inflated by earlier versions are corrected on first launch.
Hermes IDE's hermes-pty-host is a small background process that owns the PTY and its programs, starts when the first session needs it, and communicates with the UI over a token-bearing user-specific Unix socket.
Hermes IDE requires selecting the agent, model, reasoning effort, and account when creating a new task, supports saving multiple accounts per agent and reusing them as presets, and displays the cost and model used for each session.
Hermes IDE does not write the user's run logs (such as runs.tsv), nor can it prevent a vendor from retiring a model the user has chosen.
Hermes IDE launches each task in its own hermes/<name> branch and worktree; the task itself is the agent's first prompt, and the agent can be Claude Code, Codex or another CLI agent.
The author of the article is the developer of Hermes IDE.
Hermes never answers permission prompts on the user's behalf, and never types in the terminal itself.
Hermes IDE lets Claude Code, Codex, and other CLI Agents run side by side, each in its own terminal and worktree, and provides session status display, attention inbox statistics, and ⌘I/Ctrl+Shift+I to jump to the Agent that has been waiting longest.
Hermes IDE was developed by the article's author and can run command-line agents such as Claude Code, Codex, Copilot and OpenCode side by side, each running in its own terminal and worktree with its own login credentials.
Hermes IDE lets each Agent run in its own real terminal, and on macOS and Linux the Agent keeps working when Hermes exits, updates or crashes, so users can return to where they left off.
All Facts (10)
Hermes IDE's hermes-pty-host is a small background process that owns the PTY and its programs, starts when the first session needs it, and communicates with the UI over a token-bearing user-specific Unix socket.
50%UnverifiedHermes IDE does not write the user's run logs (such as runs.tsv), nor can it prevent a vendor from retiring a model the user has chosen.
50%UnverifiedHermes IDE launches each task in its own hermes/<name> branch and worktree; the task itself is the agent's first prompt, and the agent can be Claude Code, Codex or another CLI agent.
50%UnverifiedThe author of the article is the developer of Hermes IDE.
50%UnverifiedHermes never answers permission prompts on the user's behalf, and never types in the terminal itself.
50%UnverifiedHermes IDE lets Claude Code, Codex, and other CLI Agents run side by side, each in its own terminal and worktree, and provides session status display, attention inbox statistics, and ⌘I/Ctrl+Shift+I to jump to the Agent that has been waiting longest.
50%UnverifiedHermes IDE was developed by the article's author and can run command-line agents such as Claude Code, Codex, Copilot and OpenCode side by side, each running in its own terminal and worktree with its own login credentials.
50%UnverifiedHermes IDE lets each Agent run in its own real terminal, and on macOS and Linux the Agent keeps working when Hermes exits, updates or crashes, so users can return to where they left off.
50%UnverifiedHermes IDE version 2.2.9 fixed a related problem: cost and token totals kept growing while a session was open because per-session usage was counted multiple times; totals inflated by earlier versions are corrected on first launch.
50%UnverifiedHermes IDE requires selecting the agent, model, reasoning effort, and account when creating a new task, supports saving multiple accounts per agent and reusing them as presets, and displays the cost and model used for each session.
50%