S
selcukgural/imece
One terminal for many coding agents from many vendors: a leader splits the work, agents talk to each other…
View on GitHub ↗Link checked 2026-11-12 · HTTP 200 · opens the vendor's own site
- Star count
- 1 ★ · selcukgural/imece · read from the GitHub search API · (checked 2026-11-13 20:00:50)
- Source
- read directly from its GitHub page · type: open-source project
- Harvest history
- first seen 2026-11-01 · last seen 2026-11-14 · seen 14× · listed by 2 sources
- Score
- score 203 · 1 ★ · 2 sources · repo
FAQ (answers come from our records)
Who is it for?
We filed it under these tags: 开源项目. The tags come from the entry itself, not a taxonomy we imposed.
Who maintains it?
repository
selcukgural/imece;first catalogued 2026-11-01;last seen 2026-11-14。There is no official certification or quality rating here — we record sources and dates only.# imece
以下摘录自该仓库的 README(片段) · 在 GitHub 读完整 README →
One terminal for many coding agents from many vendors.
imece (Turkish: neighbours finishing a job together) runs a guild of coding
agents (Claude Code, Codex, Gemini and any agent that speaks
ACP) on one codebase, on one screen. A
leader agent splits the work into cards and hands them out; the agents talk
to each other, ask a capped pool of researchers, and lease the symbols they
change so several of them can work in the same file.
> Status: early. imece runs real Claude Code and Codex agents on one project
> (a leader that hires workers and talks to you),
> or a seeded simulation with --demo.
## Run it
sh
cd your/checkout
cargo run -p imece-tui # or, once installed: imece
cargo run -p imece-tui -- --agents 8 --budget 10 --model sonnet
cargo run -p imece-tui -- --lead codex # Codex leads; it may still hire claude workers
cargo run -p imece-tui -- --lead gemini --tier gemini:strong=gemini-2.5-pro
You do not pick a team size. lead starts alone, reads the checkout (it
can not change anything), and hires workers as the work needs them, up to
--agents (default 4). The workers, claude-1, claude-2, … each work in
their own git worktree under ~/.local/state/imece/worktrees/, detached at
your HEAD. Your checkout is never touched by an agent. When a worker sends
its done, what its worktree holds shows up in WAITING FOR YOU as a
change (Δ): enter opens the whole diff (j k scroll, [ ] file to
file), a applies it to your checkout, staged, for you to look over and
commit, r rejects it and opens the composer so you say why. Git reads
and applies changes, and moves worktrees to your HEAD, on a thread of
its own, so the screen never waits; the journal says when a change is
in. A change that does not fit (you
edited the same file, or the same lines moved on) is refused with git's
words and nothing is half applied. An applied
change is committed in the worktree as a mark (author imece), so the
worker's next done shows only what came after, and the worktree still
follows your HEAD once you commit. The trace (enter twice) shows where
each worktree is. A worktree is kept between runs: one found clean at an
older commit is moved to your HEAD when its agent next starts and between
turns; one with uncommitted changes or commits of its own stays where it
is, and the journal says so either way. The terminal's title counts what
waits for you (imece · 2 waiting), so a tab in the background shows it;
your shell's title comes back when imece ends.
Plain text goes to the lead, @claude-2 text to one agent. A message to
an idle worker opens a card; its answer closes it.
A line that starts with / is a command of the agent's own CLI, not a
message: /clear, @claude-2 /compact, /cost, /context, whatever
the vendor's command line knows. It runs as a turn and its answer comes
back like a reply. Claude Code and Gemini CLI take their commands as
they are; Codex's app server has no command line, so only /compact
works there, and /clear on a vendor without one of its own ends the
session (the next message starts a fresh one).
Every agent gets the repository's notes for coding agents, its
AGENTS.md or CLAUDE.md (at most 16 KiB), with its instructions: the
vendors' CLIs are started so that nothing of the repository loads, which
imece (Turkish: neighbours finishing a job together) runs a guild of coding
agents (Claude Code, Codex, Gemini and any agent that speaks
ACP) on one codebase, on one screen. A
leader agent splits the work into cards and hands them out; the agents talk
to each other, ask a capped pool of researchers, and lease the symbols they
change so several of them can work in the same file.
> Status: early. imece runs real Claude Code and Codex agents on one project
> (a leader that hires workers and talks to you),
> or a seeded simulation with --demo.
## Run it
sh
cd your/checkout
cargo run -p imece-tui # or, once installed: imece
cargo run -p imece-tui -- --agents 8 --budget 10 --model sonnet
cargo run -p imece-tui -- --lead codex # Codex leads; it may still hire claude workers
cargo run -p imece-tui -- --lead gemini --tier gemini:strong=gemini-2.5-pro
You do not pick a team size. lead starts alone, reads the checkout (it
can not change anything), and hires workers as the work needs them, up to
--agents (default 4). The workers, claude-1, claude-2, … each work in
their own git worktree under ~/.local/state/imece/worktrees/, detached at
your HEAD. Your checkout is never touched by an agent. When a worker sends
its done, what its worktree holds shows up in WAITING FOR YOU as a
change (Δ): enter opens the whole diff (j k scroll, [ ] file to
file), a applies it to your checkout, staged, for you to look over and
commit, r rejects it and opens the composer so you say why. Git reads
and applies changes, and moves worktrees to your HEAD, on a thread of
its own, so the screen never waits; the journal says when a change is
in. A change that does not fit (you
edited the same file, or the same lines moved on) is refused with git's
words and nothing is half applied. An applied
change is committed in the worktree as a mark (author imece), so the
worker's next done shows only what came after, and the worktree still
follows your HEAD once you commit. The trace (enter twice) shows where
each worktree is. A worktree is kept between runs: one found clean at an
older commit is moved to your HEAD when its agent next starts and between
turns; one with uncommitted changes or commits of its own stays where it
is, and the journal says so either way. The terminal's title counts what
waits for you (imece · 2 waiting), so a tab in the background shows it;
your shell's title comes back when imece ends.
Plain text goes to the lead, @claude-2 text to one agent. A message to
an idle worker opens a card; its answer closes it.
A line that starts with / is a command of the agent's own CLI, not a
message: /clear, @claude-2 /compact, /cost, /context, whatever
the vendor's command line knows. It runs as a turn and its answer comes
back like a reply. Claude Code and Gemini CLI take their commands as
they are; Codex's app server has no command line, so only /compact
works there, and /clear on a vendor without one of its own ends the
session (the next message starts a fresh one).
Every agent gets the repository's notes for coding agents, its
AGENTS.md or CLAUDE.md (at most 16 KiB), with its instructions: the
vendors' CLIs are started so that nothing of the repository loads, which
Listed badge (optional)
Copy this line into the README if you want the listing mark. Optional, and it does not affect being listed.
[](https://anclum.com/en/s/selcukgural-imece/)
- Source
- github · https://github.com/selcukgural/imece
- Browse by field
- Open source
- Traction
- 1 ★
- Listed by
- 2 independent sources
- added
- today
More like this
Filed under
Every tag opens the full list for that filter. Counts are the live catalogue figures.