Compare

Others keep agents running. Canon checks the work.

Terminals and runtimes keep your agents alive and visible. Canon puts an approved plan, acceptance criteria, an independent reviewer and a close that refuses by default in your repo, enforced by its CLI. They work side by side.

Built in Partial or different None documented, or N/A
Feature canonworkflow harness + Cockpit Herdragent runtime + clients Orcaparallel-agent workspace Warpterminal app + agents
The sprint pipeline
Plan approved before code
Yes. Ticket and plan.md, you approve first
None documented
No plan stage of its own. Its experimental Chat UI can show an agent's plan for you to approve
Agent planning and task lists, inside the app
Acceptance criteria per task
Yes. acceptance.md, every box checked to close
None documented
A line its orchestration skill tells a coordinating agent to write. Orca does not check it
None documented
Review by an agent with no build history
Yes. Fresh evaluator, Read and Bash only
None documented
No built-in reviewer agent. You can set up a review-only worker or a scheduled review prompt. Its diff review is a human annotating
PR review agents in Factories
Close blocked by a mechanical gate
Yes. The CLI refuses on an unchecked box, missing summary or non-pass verdict. Bypass needs an explicit flag or a recorded waiver
None documented
Experimental decision gates, run from its CLI, block a task until resolved. It also shows GitHub's own checks
Approval gates in a Factories config
Evidence behind each verdict
Yes. A file:line cite per criterion
None documented
Stores a worker's report. It does not verify the evidence behind it
Screenshot or video artifacts
Plan-vs-actual receipt
Yes. One row per criterion: delivered, waived, deferred or partial
None documented
Its coordinating agent is told to report each task's outcome and evidence. No comparison with a plan found
None documented
Decisions kept in your repo
Yes. .tickets/ markdown, read back at the next start
Sessions persist, decisions are not tracked
It does not write plans or decisions to your repo. It reads config and templates, such as orca.yaml
Conversations stored locally, optional cloud sync
Grades PRs in CI
Yes. sprint-headless runs reviewer, evaluator and security review
None documented
Scheduled automations can run a prompt such as a PR review, but not on a PR event. It also shows PR check state
PR review in Factories
Running agents
Work survives closing the UI
Browser tab yes, a daemon owns the terminals. If the daemon stops, one click resumes
Yes. The server owns the terminals
Yes. A background daemon owns the terminals
Local and cloud-synced conversations
Agent state at a glance
Yes. Working, idle or needs you, with desktop notifications
Yes. Blocked, working, done, idle
Yes. Working, waiting on you, done, blocked
Conversation panel and agent notifications
Multi-machine and SSH
Local only today
Yes. Saved SSH machines in one window
Yes. SSH worktrees and remote servers, plus a mobile app in beta
Cloud agents in Factories
Agents supported
Claude Code, Pi and Copilot CLI in the Cockpit
22 agent CLIs detected
34 agent CLIs named in its README, and any CLI agent
Claude Code, Codex, OpenCode and its own agent
Project and ticket board
Yes. A kanban per project, built from .tickets/
None documented
A board of worktrees by status. Tickets live in GitHub, GitLab, Linear or Jira, not in your repo
Task lists for agent work only
Repo health checks
Yes. The context-check, context-doctor and dead-code-cleanup skills, run in a session, report on your agent context and unused code; dead-code-cleanup removes code only after you confirm
None documented
No check of your agent instructions, context files or dead code. It offers a scheduled "Repo health" prompt template
None documented
Runs in your existing terminal
Cockpit is a browser window. The CLI works in any terminal
Yes. TUI, CLI and plain SSH
Its own terminals in an app. Its CLI works from any shell
Its Terminal replaces yours. Its Agent CLI works in any terminal
Ownership
Local-first, no account
Yes. No SaaS, nothing hosted
Yes. Herdr Cloud is coming soon
The base app needs no account. An optional Orca account is needed to share artifacts, publish skills and pair a phone over its relay
Terminal downloads freely. Factories needs signup
License
MIT
Apache 2.0
MIT
Terminal is open source. Factories and Agent CLI licensing not stated
Cost
Free. You pay your model provider
Free
Free and open source. You bring your own agent subscription
Factories is usage-based per agent run

Checked 2026-10-01 against each product's own pages. Orca was read at its docs and source (commit 42a40f8) on 2026-10-07. Canon's gate was run against a scratch ticket, and it refused on a missing summary and on empty acceptance checklists. "None documented", "N/A" and cells that say Orca does not do something describe what we found in its docs and source on that date, not that it cannot exist. Next check due 2026-12-31.

In practice

Where each one wins.

Canon doesn't replace a runtime or a terminal. It answers a different question: is the work done?

vs Herdr

Herdr keeps the herd alive. Canon decides what counts as done.

Herdr is a runtime: a server owns the terminals, agents keep working while you are away, and it tracks whether each one is blocked. Canon's Cockpit does a lighter version of that for one machine. What Herdr does not document is a plan, acceptance criteria or a review gate.

Herdr wins on multiple machines, SSH and the number of agent CLIs it detects. Use both: canon's CLI and skills run inside any terminal, including a Herdr pane.

vs Orca

Orca runs a fleet of agents. Canon decides what counts as done.

Orca is a parallel-agent workspace: its own terminals and diff view, worktrees with a board of their status, a status for each agent, SSH and a mobile app. Its diff review is a human annotating a diff and sending the comments to an agent, and its orchestration layer is experimental and opt-in. What we found no sign of in Orca is a plan stage of its own, a built-in grader or an enforced acceptance check. Canon puts a plan, a checklist and a fresh grader between "the agent says done" and done.

Orca wins on parallel agents across machines, the number of agent CLIs it names, diff review inside the app and a mobile app. Use both: canon's CLI and skills run in any terminal, which includes Orca's.

vs Warp

Warp moves the work into its window. Canon keeps it in your repo.

Warp pairs a terminal with agents, planning and code review, and Factories adds team orchestration, approval gates and run scoring across Slack, Linear, Jira and GitHub. Canon stays local: plans and verdicts are markdown files in your repo, graded by a second agent, enforced by the CLI.

Warp wins on team workflows, integrations and cloud agents. Use both: canon's sprint runs under Claude Code, which Warp can also run.

Get started

Keep the terminal you have.

Install once, register the sprint skill, then ask your agent to run sprint start.