Config-driven agents
Define agents with prompts, tools, and models in YAML—check into git for team sharing.
Continue is an open-source extension for VS Code and JetBrains that combines chat, autocomplete, and custom agents defined in `config.yaml`—a lightweight way to package skill-like workflows.
Open ContinueAdd Continue from the marketplace; open the Continue panel in VS Code or JetBrains.
Add API keys or local model endpoints in config—never commit secrets.
Create named agents with system prompts and allowed tools for one workflow each.
Store config in the repo (minus secrets) so everyone uses the same agent definitions.
Select the agent that matches your task (fix, review, docs) from the Continue UI.
Reference files, docs, or git diff so the agent stays precise.
Use bundled commands for repetitive checks your team defined.
When output drifts, update the agent prompt or rules in YAML and PR it.
Define agents with prompts, tools, and models in YAML—check into git for team sharing.
Pull rules from files and `@` context providers so each agent stays scoped.
Run local models (Ollama) or cloud APIs from one extension surface.
Quick-invoke packaged workflows from the chat input—handy for reviews and doc updates.
Commit `config.yaml` agents for “security review” and “write tests.”
Use local models when code cannot leave the network.
Different agents per language with rules in each package.
The extension is open source; you still pay for cloud model APIs or run local models yourself. There is no vendor-hosted agent runtime like Copilot’s cloud.
Continue emphasizes YAML-defined agents and model choice in an extension. Cline and Cursor ship deeper autonomous agent UX; many teams use Continue for configurable, git-tracked agent presets.
Continue’s tool surface has expanded over time—check current docs for MCP or built-in context providers versus pairing with external tools.
Note: Continue config formats evolve—pin extension versions in team docs and read continue.dev/docs when upgrading.