Delv
CodingActive· 5dby Continue★ 4.3

Continue

Open-source AI coding agent for VS Code and JetBrains with chat, plan and agent modes and full model and deployment flexibility.

Try it Source Also in Tools: AI Code & Dev →
B
Safety & Trust

Delv Safety Grade: B

Score 72/100 · assessed 2026-04-18

Maintainer75
Permissions45
Supply chain80
Transparency90
Incidents100

Continue is a well-maintained open-source coding agent from a venture-backed startup with active development and strong community engagement. The project demonstrates excellent transparency with comprehensive documentation, public roadmap, and responsive issue tracking. However, as an autonomous coding agent, it requires extensive permissions including filesystem write access, shell execution capabilities, and network access to external LLMs. The agent can modify code, execute commands, and interact with your development environment with minimal sandboxing. Supply chain is solid via standard extension marketplaces (VS Code, JetBrains) with versioned releases. The freemium model with optional paid hub services is clearly documented. No known security incidents, but the broad permissions inherent to AI coding agents warrant careful consideration of what codebases and credentials are exposed.

Green flags

  • Fully open source with 13k+ GitHub stars and active maintenance
  • Distributed via official VS Code and JetBrains marketplaces
  • Comprehensive documentation and transparent development process
  • Supports self-hosted models for data sovereignty
  • Active community with 400+ contributors

Red flags

  • Autonomous agent with shell execution and filesystem write access
  • Can read environment variables including potential secrets
  • Network access to external LLMs means code may leave your machine
  • Limited sandboxing for agent actions in development environment

Permissions requested

Read filesWrite filesShell executeOutbound networkRead envAccess secretsExternal LLM callRepo readRepo write
Assessed by Delv Editorial using public metadata. Grades are advisory and update as the ecosystem changes. They do not replace your own review of permissions and code before granting an agent access to sensitive systems.

Pricing

FREEMIUM — Free OSS, paid hub

Platforms

vscodejetbrainscli

Review

Continue sits in the awkward middle ground between Copilot's autocomplete and full agent frameworks like Aider. It gives you three modes: chat (ask questions, get answers), plan (outline changes before applying them), and agent (autonomous multi-file edits). The agent mode is where the autonomy claim lives, but it's more supervised than you'd hope. You still approve each step, which makes it less of an agent and more of a chatbot with a checklist. I've used it most for refactoring across multiple files. You describe the change, Continue generates a plan with file paths and line numbers, then you approve or reject each edit. It's faster than manual search-and-replace, slower than trusting Aider to just do it. The planning step catches mistakes before they land in your codebase, which matters if you're working on something you can't afford to break. But it also means you're babysitting every decision. The model flexibility is the real draw. You can plug in Claude, GPT-4, local Ollama models, or your company's air-gapped deployment. That's rare. Most coding agents lock you into one provider. If you're in a regulated industry or just want to avoid sending code to OpenAI, Continue is one of the few options that doesn't feel like a compromise. Autocomplete works but isn't better than Copilot. Chat is fine for explaining code or debugging, though it doesn't remember context as well as Cursor. The agent mode is where it tries to differentiate, but the approval loop makes it feel like a power tool with training wheels. I'd reach for it when I need model flexibility or when I'm refactoring something large enough to benefit from a plan but too risky to let an agent run wild. The open-source angle matters if you want to self-host or extend it. The paid hub adds team features and analytics, but the core agent is free. That pricing model works if you're a solo developer or a small team that doesn't need centralised billing. Larger orgs will want the hub. Compared to Cursor, it's less polished but more flexible. Compared to Aider, it's more cautious but easier to adopt in an IDE.
Verdict

Best for teams that need model flexibility or air-gapped deployments. If you just want fast autocomplete or a truly autonomous agent, look elsewhere. The planning mode is useful for large refactors, but the approval loop limits how much time it actually saves.

Good at

  • Works with any model: Claude, GPT-4, Ollama, self-hosted
  • Planning mode shows changes before applying them
  • Open-source core with no vendor lock-in
  • Supports air-gapped and on-premise deployments
  • Available in VS Code, JetBrains, and CLI

Watch out

  • Agent mode requires approval for every step, limiting autonomy
  • Autocomplete doesn't match Copilot's quality
  • Context retention weaker than Cursor
  • Paid hub needed for team features
  • Planning overhead slows down small tasks

Use cases

  • autocomplete
  • chat coding
  • air-gapped coding