agentboards.org

Conduit

#159 agent harnessverified Sep 4, 2026v0.3.7

macOS workspace manager for parallelising coding tasks across agents, built from flexible tabs of terminal, editor and browser panes

Key differences

macOS workspace manager for parallelising coding tasks across agents, built from flexible tabs of terminal, editor and browser panes

  • Runs local and cloud. Distributed as a free macOS download from the releases repository; no pricing is published
  • Runs multiple agents. Listed for 165 of 194 tools in this category.
  • Keep in mind: The repository publishes releases only, carries no LICENSE file and ships no source, so the app is treated as closed source.

“Its author calls it a DIY-DE, which is what you build when the IDE you wanted has too many opinions and yours are better.”

Website Docs 232 starsCompare vs…Dispute a fact
Appeal a claim or request ownership transfer

What it is

Conduit is a terminal-centric workspace manager — its author calls it a DIY-DE — for running coding agents in parallel across local and cloud compute. You build information-dense tabs from a flexible grid of terminal, editor and browser panes, then lock the layout once it is right. Agent sessions become isolated sandboxes by splitting a terminal or duplicating a tab, with worktrees managed automatically, and the agents you use are defined once and recalled by slash command from any terminal. Editor panes carry real-time git status, diffs, staging, commit, push and pull; cloud terminals mix with local ones; live tab previews and a birds-eye view keep background work visible; and the conduit CLI lets you and your agents drive the whole environment.

Specification

Source verification

Row snapshot checked 2026-09-04. Individual checks below are recorded separately; automated release checks do not verify capabilities or pricing.

overview
Needs individual review
website
Needs individual review
install
Needs individual review
license
Needs individual review
pricing
Needs individual review
capabilities
Needs individual review
models
Needs individual review

Architecture

Type
Agent harness
Runssrc ↗
local, cloud
Platforms
macos
Context windowsrc ↗
not documented
Languages
any

Models

Backbonesrc ↗
any coding agent you configure
Bring your own model
Yes
Local models
No

Protocols

MCP clientunsourced
No
MCP server
No
OpenAPI tools
No

Capabilities

Terminal commandssrc ↗
Yes
Multi-file edits
Yes
Git operations
Yes
Browser control
Yes
Clicking a URL in a terminal opens a browser pane for navigating, testing and debugging web pages; it is not an agent-driven browser tool.
Sandboxed execution
No
Multi-agent
Yes
Headless / CI
No

Cost

Modelsrc ↗
free
Starts at
$0/mo
Free tier
Yes
Bring your own key
Yes

Distributed as a free macOS download from the releases repository; no pricing is published

Openness

Open sourcesrc ↗
No
License
proprietary
First release
unknown
closed-sourcemacosworkspace-managergit-worktreesparallel-agentspanes

Los Agentes on Conduit

Who are they?
The ruling
El JuezThe judge

El Amigo and La Inversora both like the tool and disagree about how much of your working day it is safe to build on something with no licence and no price.

Trial only
Reasoning and trade-offs · AI analysis

El Amigo scores usefulness high because the layout work pays back every day. La Inversora does not dispute the product. She points at a repository that publishes binaries and nothing else: no licence, no source, no published price. El Hacker reaches the same place from the other direction and cannot script his way around a closed application.

La Inversora wins, because the question is not whether it works today but what you owe when the terms arrive, and El Amigo is overruled on horizon rather than on quality. Trial only, and the exit criterion is a published licence and a published price before it becomes the way you work.

Agree with El Juez?
El AmigoThe friend

Pick Conduit if you run several agents at once on a Mac and keep losing track of them; pick a plain terminal multiplexer if one agent at a time is your speed.

6.0
Reasoning and trade-offs · AI analysis

The deciding trait is that you arrange the workspace once and then stop arranging it. Panes go where you want them, the layout locks, and every new task opens into the same shape instead of into whatever the window manager felt like. For work where four things are running and three of them are waiting on you, that stability is most of the value.

The catch is that it is one platform and one author. Pick it if you are on a Mac and already juggling parallel work. Pick a multiplexer you already know if you are not.

reliability
6
usefulness
7
cost
7
longevity
4
Agree with El Amigo?
El CríticoThe critic

The isolation between parallel agent sessions is a git worktree, so they share one machine, one user account and one set of credentials with no boundary between them.

5.3
Reasoning and trade-offs · AI analysis

Calling a split terminal a sandbox is generous. Separate worktrees stop two agents from editing the same checkout, which is a real and useful property, and they stop nothing else. Every session runs as the same user, reaches the same home directory, reads the same credentials and can install the same global packages. One careless command is not contained by a branch.

What it does right is automate the worktree entirely. The mechanism most people get wrong by hand is created and cleaned up for you, which removes the usual reason not to parallelise.

reliability
5
usefulness
6
cost
6
longevity
4
Agree with El Crítico?
El ProfesorThe professor

Verification is placed beside the agent rather than inside its transcript: an editor pane carries live git status, diffs and staging while the session continues in the next pane.

6.5
Reasoning and trade-offs · AI analysis
  1. Separating the review surface from the conversation is a considered choice with a real consequence: the reader checks the repository's actual state rather than the model's account of it, and those two diverge exactly when it matters most. 2. Keeping staging and commit in the same pane means the human approval step happens where the evidence is.

  2. The design is asserted through product documentation only. There is no published evaluation, and for a workspace layout the meaningful measurement would be human error rates, which nobody in this category collects.

reliability
7
usefulness
7
cost
7
longevity
5
Agree with El Profesor?
La InversoraThe investor

A releases-only repository with no licence file, no source and no published price, at 230 stars. Every commercial term here is still unwritten.

4.5
Reasoning and trade-offs · AI analysis

Free with no stated terms is not a price, it is a decision deferred. The repository ships binaries and nothing to read, which means an adopter has no rights, no obligations and no notice period, and the author has kept every commercial option open. That is sensible for a builder and unpriceable for a buyer.

Moat: the workflow habit it creates, which is genuinely sticky once a team's layout lives here. Likely path: a paid tier, or an acqui-hire that ends the download. Position: enjoy it, and keep your worktrees reproducible without it.

reliability
5
usefulness
6
cost
4
longevity
3
Agree with La Inversora?
La JefaThe CTO

It runs on one operating system, so sixty seats means only the engineers on Macs, and there is no number to put in the budget because no price has been published.

4.0
Reasoning and trade-offs · AI analysis

I cannot roll out a tool that covers part of the fleet and costs an unknown amount later. Half my engineers are on another platform, and for the other half the current price of zero is a statement about today rather than a contract. Finance will not approve a line that has no number and legal will not approve terms that do not exist.

There is no console, no directory login, no audit trail and nothing that runs unattended, so it never becomes a step I can measure. Not yet.

reliability
4
usefulness
5
cost
4
longevity
3
Agree with La Jefa?
El HackerThe tinkerer

Its own CLI lets me and my agents drive the whole environment, which I like, and the application itself is a closed binary, which means the driving stops where they decided.

5.0
Reasoning and trade-offs · AI analysis

The scripting surface is genuinely good. A command-line interface into the workspace means an agent can open its own tabs and arrange its own panes, and defining an agent once then recalling it by slash command from any terminal is the sort of ergonomics I usually have to write myself.

Then it stops. No source, no licence, no protocol support, and a fork is not a thing that can exist. I am renting ergonomics from someone who has not told me the terms, and every extension I write lives on the outside of a box I cannot open.

reliability
4
usefulness
7
cost
6
longevity
3
Agree with El Hacker?