Paradev

Detailed report

Paradev documents headless Linux execution on your own servers and control from other devices. Remote access requires a paid plan and the vendor relay. Recent reports identify remote-session problems worth testing.

Product and evidence

The product is Paradev, formerly Superset, maintained at ParadevAI/Paradev. The official rename announcement is dated October 26, 2026. This is not Apache Superset. Undated documentation below was checked on November 13, 2026. That is an access date, not a publication date.

What it offers

Paradev lets you keep your preferred agents, subscriptions, repositories, and hardware while putting parallel work, monitoring, review, and remote control in one workspace. A useful scenario is running agents on an always-on Linux server, switching between laptops, and checking results from an iPhone without manually rebuilding terminal and branch context. Its MIT-licensed source and standard Git worktrees offer a credible exit path. It reduces the operational work around agents. It does not establish that running 100 agents produces better software. Official overview, repository.

Agents, models, and payment

Documented agents include Claude Code, Codex, OpenCode, Cursor Agent, Copilot, Gemini, Amp, Pi, Oh My Pi, Droid, Kimi Code, Mistral Vibe, Grok, and others. Integration depth varies: Claude and Codex have status reporting, session restoration, usage history, and subscription quota meters. Some other agents run mainly as terminal programs. Model selection depends on the underlying agent and account, rather than a universal Paradev entitlement. Support matrix.

You bring existing subscriptions or API keys. Paradev says it does not proxy model calls. OpenRouter, AWS Bedrock, Google Vertex, and Vercel AI Gateway configuration is documented, with agent-specific settings. Do not assume every provider works identically across terminal, chat, and managed cloud: cloud workspaces use account sign-ins and explicitly ignore several credential environment variables. Provider configuration, cloud credentials.

Your machines versus vendor cloud

Own-server support is explicit: Linux x64 and arm64, a standalone CLI, installed agent CLIs, Git/GitHub authentication, and an API key for unattended login. A documented systemd configuration keeps the host service available across reboots. Repositories and agent sign-ins reside on that server. Bring Your Own Cloud.

SSH can bootstrap a host from the desktop interface. Subsequent application traffic uses Paradev Relay, rather than a documented persistent SSH connection. The host binds to loopback and needs no inbound ports. Other authorized devices can use its terminals, agents, and diffs. Desktop port forwarding carries TCP, supports local port remapping, and follows the selected workspace. UDP is unsupported. This is remote access to a workspace that remains on its host, not documented automatic migration between laptops. Remote architecture.

Managed cloud is separate: Paradev provisions sandboxes from environments containing repositories, variables, and setup configuration. Agents can continue while your laptop is closed. Those environments require GitHub integration and Team or Ultra, whereas your own Linux host uses paid remote access. Cloud workspaces.

Session recovery limits

A background terminal daemon preserves processes and scrollback through ordinary app quits and updates. However, quitting the desktop app stops its host service and remote reachability unless you use the standalone CLI host. A reboot ends processes. Supported agents can restore conversations using their session IDs. That is recovery, not uninterrupted execution. A sleeping laptop cannot keep computing. Keep Agents Running, session recovery.

Workflow and orchestration

Each task can receive a branch, Git worktree, and terminals. Worktrees separate working files but do not sandbox processes or prevent integration conflicts. Setup scripts remain necessary for dependencies and environment files. Repository explanation, troubleshooting.

The superset:orchestrate skill lets Claude Code or Codex coordinate mixed-agent workers through CLI commands: create workspaces, launch bounded tasks, read terminals, send follow-ups, and collect results. Scheduled and event-triggered automations are also documented. Tests run through agents or project commands. The orchestration layer does not guarantee acceptance criteria were met. Orchestration mechanics.

Review tools include staging, commits, pushes, PR creation, GitHub checks, review threads, and sending selected diff lines to an agent. These features make parallel output easier to assess, but human review and repository CI remain essential. Diff viewer, PR workflow.

Setup, security, and teams

Desktop onboarding detects prerequisites and installed agents. Headless setup requires more administration. Own-server execution is documented, but I found no complete supported recipe for privately hosting the entire account, relay, and billing infrastructure. MIT licensing should not be confused with a ready-to-use, vendor-independent service deployment. Installation, host server.

Remote terminal and port traffic passes through the vendor relay, while agents send context to their configured model providers. The reviewed documentation does not establish end-to-end encryption. Host access is broad: Paradev warns that authorized users can reach what the host can, including agent sign-ins. Teams should use dedicated hosts and intentionally scoped credentials. Enterprise advertises SSO, SCIM, audit logs, and a SOC 2 Type II report. These are vendor claims requiring procurement review. Own-host security, Enterprise scope.

Pricing checked November 13, 2026

All prices are USD and exclude agent subscriptions/API charges:

• Free: one user, local workspaces, desktop app, CLI.
• Pro: $20/user/month, or $192/year ($16/month equivalent). Includes remote access, mobile, automations, Slack and Linear integrations.
• Team: $60/user/month, or $648/year ($54/month equivalent). Adds managed cloud, five environments, and 1,200 GB-hours per seat monthly.
• Ultra: $200/user/month. Includes ten environments and 4,000 GB-hours per seat monthly.
• Enterprise: custom, annual billing.

Cloud pools are shared across the team. Advertised 150/500 hours assume an 8 GB machine. Larger machines consume the pool faster. Additional usage costs $0.06/GB-hour. Pro is the relevant starting tier for your own servers. Pricing.

There is documentation drift: the August 9, 2026 changelog says remote hosts became available on every plan, while current pricing and setup docs require paid access. Budget from current pricing and verify entitlement during trial. Dated changelog.

Development and user reports

Desktop 2.0.8 was released November 10, 2026, with a November 12 canary. This shows active development, not proven stability. Releases.

Independent sentiment is sparse. An April 22, 2026 Reddit discussion contains a user calling Superset “a game changer,” without enough detail to establish why or generalize satisfaction. Founder recommendations and vendor testimonials are not independent validation. Discussion.

More useful negative evidence comes from specific reports:

• November 9: chat on a CLI-installed host failed because an adapter package was missing. #9560.
• November 10: remote OpenCode showed models unavailable on the host. #9597.
• November 12: closing ports reportedly also killed agent sessions. #9683.

These are reported failures, not measured prevalence. Automated issue-analysis comments are not additional independent reproductions.

Fit and trial

Choose Paradev if you want mixed agents, your own always-on compute, and integrated review. Avoid it if you require production-supported Windows, Android, browser-only control, or fully private remote infrastructure. Teams gain shared hosts and automation, but broad credential access deserves attention.

Trial Pro on one dedicated Linux server: run two independent tasks with different agents, switch laptops, reconnect after a network interruption, review from mobile, restart the host, and complete a PR through CI. Test your actual providers and large repository. Assess recovery, reliability and review effort.

Platform facts: installation, FAQ.