Sokudo vs Paradev
Paradev, formerly Superset, and Sokudo both run coding agents in parallel worktrees. Sokudo adds the rest of the workplace: agent personas across providers, a Kanban board that carries dependencies and a reviewer step, a Wiki Engine, a knowledge base, rules, and memory, and a harness that learns from delegated runs and proposes rules for your approval.
Reviewed · Published by Sokudo · Sources and scope
Pick Sokudo
When agents need roles, review, and a shared memory.
Give each agent a persona and a provider, then delegate Kanban tickets with dependencies to implementation agents and a reviewer. Every agent draws on the Wiki Engine, knowledge base, rules, and memory, and delegated runs can save lessons and propose rules you approve. Design on the canvas, query the database, and work over SSH beside the code.
Pick Paradev
Parallel agent sessions you can reach from anywhere.
Paradev runs each agent in its own worktree with terminals, diff and pull request review, and a browser pane. It adds automations, cloud workspaces, an iPhone and iPad app, a CLI, an MCP server, and an early-alpha TypeScript SDK. Its docs do not describe a built-in codebase wiki or a Paradev-managed knowledge base, rules, or memory.
Where Sokudo pulls ahead.
Both run agents side by side with terminals, change review, a browser, and remote machines. The difference is what surrounds the work. Here is what each product documents.
| Criterion | Sokudo | Paradev |
|---|---|---|
| What you get | An IDE with agents, design, data, and browser tools in one workspace. | An open-source workspace that runs CLI coding agents in parallel, each on its own branch, locally or in the cloud. [2] |
| Orchestration | Delegate tickets across providers, with implementation agents in their own worktrees and a reviewer. | An orchestrate skill turns Claude Code or Codex into a coordinator that starts worker agents in separate workspaces and checks their reports. [7] |
| Personas | Custom agent roles, instructions, and model choices. | Settings for each agent: label, launch command, and task prompt template, with model pickers for built-in agents, plus custom agents for any CLI. Separate roles that share one CLI, such as an implementer and a reviewer, are not described. [6] |
| Work tracking | Kanban tickets, dependencies, and worker review in the IDE. | A Tasks view for Linear issues, GitHub issues, and Paradev tasks, in list or board layouts, that starts an agent in a new workspace for each task. [8] |
| Wiki Engine | Citation-backed Pages and an interactive Graph of your codebase, read by you and your agents. | A generated codebase wiki is not described in Paradev’s docs. [5] |
| Project context | A workspace knowledge base, rules, and memory that agents draw on in later sessions. | Installs Paradev’s skills, and the skills and tools of plugins you add, into each agent’s own configuration, and the agents may keep their own instruction files. A Paradev-managed knowledge base, rules, or memory store is not described. [10] |
| Self-improving harness | Finished runs can save learnings to the knowledge base and propose rules that wait for your approval. | A post-run learning loop is not described in Paradev’s docs or changelog. [11] |
| Design canvas | Design screens and hand the canvas to agents in the same workspace. | Pages: agents publish HTML designs and reports, teammates pin comments, and an agent revises the page. [12] |
| Database | Inspect schemas and query data in the database manager. | Plugins connect agents to hosted databases, such as Supabase for tables, SQL, and migrations. A database client in the app is not described. [13] |
| Remote work | Remote SSH sessions and commands alongside the workspace. | Remote hosts set up over SSH and reached through Paradev’s relay, plus cloud workspaces that run without your own computer connected. [14] |
| Browser | An in-app browser with extension support for previews and wallet flows. | A browser pane with DevTools, Chrome extensions, and Design Mode, which sends a clicked element to an agent. Agents can drive it from the CLI. [15] |
“Not described” means the cited Paradev pages do not document an equivalent feature; it does not prove the feature is impossible. Some cited Paradev features need a paid Paradev plan. Sokudo capabilities: [1].
Open Sokudo when the work needs the whole workplace.
Choose Sokudo if
- You delegate to named agents with distinct roles, instructions, and models.
- You want tickets, dependencies, implementation, and review on one Kanban board.
- You want every agent to start from a shared codebase wiki, knowledge base, rules, and memory.
- You want finished runs to save learnings and propose rules for you to approve.
- You want a design canvas and a database manager built into the IDE.
- You want SSH, app previews, and compatible browser extensions beside the board and the knowledge.
Choose Paradev if
- You want automations that start agents on a schedule or from GitHub, Slack, Linear, or Sentry events.
- You want cloud workspaces and an iPhone or iPad app to keep agents going away from your desk.
- You want to script workspaces and agents from a CLI, an MCP server, or an early-alpha TypeScript SDK.
- You review and merge GitHub pull requests in the app and send diff lines back to an agent.
- You want to share agent-built reports with your team as Pages and have an agent act on comments.
- You want MIT-licensed source you can inspect and modify.
What you actually get.
Straight answers about Sokudo and Paradev.
What is the difference between Sokudo and Paradev?
Paradev, formerly Superset, and Sokudo both run coding agents side by side in separate worktrees, with terminals, change review, an in-app browser, and remote machines. Paradev focuses on running and reaching those sessions: automations, cloud workspaces, an iPhone and iPad app, Pages, a CLI, an MCP server, and an early-alpha SDK. Sokudo builds the workplace around the agents: custom personas, a Kanban board that carries dependencies and a reviewer step, a Wiki Engine, a knowledge base, rules, and memory, a design canvas, and a database manager.
Is Paradev the same product as Superset?
Yes. Superset, the desktop app for running coding agents in parallel, renamed itself Paradev in October 2026. Its website, docs, and GitHub repository now redirect to paradev.com and ParadevAI/Paradev, and its command-line tool is still called superset. It is unrelated to Apache Superset, the business-intelligence tool.
Can both tools orchestrate agents?
Yes. Paradev’s orchestrate skill turns Claude Code or Codex into a coordinator that starts worker agents in separate workspaces through its CLI and checks their reports before you integrate the results. Sokudo delegates Kanban tickets to a roster of agents with roles, instructions, and models, tracks dependencies between tickets, and keeps implementation and review as separate steps.
What is the Sokudo Wiki Engine?
The Wiki Engine maps your project into citation-backed Pages and an interactive Graph of files, dependencies, and modules. You use it to understand the architecture and how modules fit together. Your agents can search and read the same pages, so people and agents work from one shared guide to the codebase.
How does Sokudo keep project context for agents?
Each Sokudo workspace has three context stores. The knowledge base holds reference material such as architecture notes and conventions. Rules hold the instructions agents should follow. Memory holds durable decisions, preferences, and gotchas, found by meaning rather than keywords. Agents draw on all three in later sessions. Paradev installs its own skills, and those of plugins you add, into each agent’s configuration so agents can drive Paradev and connected services, and the agents may keep their own instruction files; its docs do not describe a Paradev-managed knowledge base, rules, or memory store.
How does self-improvement work in Sokudo?
When self-improvement is on for a workspace, Sokudo reviews each delegated run after its agents finish. When a run produces something worth keeping, Sokudo saves it to the knowledge base or proposes a rule, and marks the entry as written by an agent. A proposed rule does not take effect until you approve it, and you can turn the feature off for any workspace. It changes the context your agents receive; it does not retrain any model.
Where do databases fit in each tool?
Sokudo’s database manager is built in: connect PostgreSQL, MySQL, SQLite, MongoDB, Redis, ClickHouse, and more, inspect schemas, and ask agents to prepare queries, then review proposed changes before they are committed. In Paradev, agents reach hosted databases through plugins such as Supabase and Neon, connected with your own account; its docs do not describe a database client built into the app.
When should I choose Paradev?
Choose Paradev when you want agents started by schedules and events, cloud workspaces, an iPhone or iPad app, a scriptable CLI and MCP server, or MIT-licensed source. Choose Sokudo when you want named agents, tickets with dependencies and review, shared project knowledge that grows with approved learnings, a design canvas, and a database manager in one workplace.
The workplace, not the window.
Give each agent a persona. Keep the board, the canvas, the database, the remote machine, and the browser in Sokudo.