Continue is an open-source (Apache 2.0) tool that has evolved from a VS Code / JetBrains autocomplete extension into a broader source-controlled AI platform. In 2026 its flagship feature is markdown-authored AI checks that run in CI and surface as GitHub status indicators on pull requests, plus a cn CLI (the open-source Continue CLI) for local development and automation.
Continue in 2026 is primarily a CI + source-controlled-checks tool. The IDE extension still ships,
but the center of gravity has moved to .continue/checks/ and the cn CLI.
The IDE extensions still exist; this page focuses on the CLI + CI surface because that's where Continue is strongest today.
Prefer the npm install path on untrusted networks — curl | sh executes whatever the server
returns. If you use the script, inspect it first by opening the raw URL in a browser before piping
it to a shell.
# macOS / Linux — install script
curl -fsSL https://raw.githubusercontent.com/continuedev/continue/main/extensions/cli/scripts/install.sh | bash
# Or via npm (requires Node 20+)
npm i -g @continuedev/cli
# Check
cn --versionIDE extensions: install Continue from the VS Code Marketplace or JetBrains Marketplace.
The core pattern: AI review criteria are markdown files in your repo, not dashboard config. Add .continue/checks/ to your repo:
your-project/
├── .continue/
│ └── checks/
│ ├── api-design.md
│ ├── no-todos.md
│ └── test-coverage.md
└── src/Example check:
---
name: no-todos
description: Block PRs that add new TODO/FIXME comments to production code
---
Review the diff. If it adds any new `// TODO` or `// FIXME` comments in files
under `src/` (excluding tests), fail the check with a message pointing to the
offending lines. Otherwise pass.On each PR, Continue runs every check and posts pass/fail as a GitHub status with suggested fixes for failures. Checks are versioned with your code — reviewers see the exact criteria that ran.
cn CLIcn is the command-line entry point — an interactive TUI agent for local development, plus a headless mode (-p) for scripting and CI:
# Interactive TUI session
cn
# Headless mode — single-shot, no TTY needed (CI, Docker, scripts)
cn -p "Generate a conventional commit message for the staged changes"
# Pick a model from the hub (repeatable)
cn --model anthropic/claude-sonnet-4-6
# Use a specific config / assistant
cn --config ~/.continue/config.yaml
# Resume the last conversation
cn --resumeHeadless mode supports piped input and JSON output, so cn slots into CI pipelines and shell scripting (--allow / --ask / --exclude gate which tools the agent may run).
Continue is model-agnostic. Supported via config: Anthropic, OpenAI, Google, Ollama, Groq, Mistral, and others. Configure via .continue/config.yaml (or config.json in older setups) with your API keys.
Add a minimal check to your repo:
mkdir -p .continue/checks
cat > .continue/checks/no-secrets.md <<'EOF'
---
name: no-secrets
description: Block PRs that add hardcoded API keys or tokens
---
Review the diff. If it introduces any hardcoded strings that look like API keys
(OpenAI sk-, Anthropic sk-ant-, GitHub ghp_, AWS AKIA, or generic base64
secrets >30 chars), fail the check with the offending file and line numbers.
EOF
git add .continue/
git commit -m "add no-secrets AI check"
git pushEnable the Continue GitHub App on your repo, open a PR — the check runs automatically.
cn vs IDE extension confusion. They share some config but aren't identical. If something works in one and not the other, check the docs for each.