Continue
Type: full-code · Vendor: Continue Dev, Inc. · Language: TypeScript · License: Apache-2.0 · Status: active · Status in practice: mature · First released: 2023-05-24
Open-source AI dev tooling that started as an in-IDE coding assistant (Chat/Edit/Agent/Autocomplete in VS Code and JetBrains) and has pivoted to a CI-enforceable 'AI checks on every PR' framing via the `cn` CLI; the legacy IDE agent remains shipped.
Description. Continue today positions itself around the `cn` CLI that runs AI checks on pull requests. Each check is a markdown file in `.continue/checks/` whose result surfaces as a GitHub status check (green/red, with suggested fix). Underneath, the same engine still powers the long-standing VS Code and JetBrains extensions that provide Agent mode (autonomous task completion with tools and MCP), Chat (Q&A), Edit (inline edit), and Autocomplete (predictive completions). Custom assistants/blocks can be configured through Hub configs and via `mcpServers` for MCP integration.
Agent loop shape. Two distinct loops live in the project: (1) PR-check loop — on every PR, each check's markdown prompt runs against the diff and results post as GitHub status checks; failures suggest a diff users can accept from GitHub. (2) IDE agent loop — a ReAct tool-use loop where the model uses chat-equipped tools (file edit, terminal, MCP, codebase context), asking permission per-tool by default; tool outputs feed back into the model as context. Agent mode requires MCP tools; MCP is only available in agent mode.
Primary use cases
- in-IDE agent + chat + edit + autocomplete (VS Code, JetBrains)
- PR-check enforcement via `cn` CLI + GitHub status checks
- MCP-tool extensibility inside Agent mode
- Hub-shared assistant / block / MCP configurations
Key concepts
- AI checks (`cn` CLI) (docs) — Markdown checks in `.continue/checks/` enforced in CI via GitHub status checks.
- Agent mode (IDE) → react (docs) — Autonomous tool-using agent inside VS Code/JetBrains.
- MCP → mcp (docs) — Only usable in agent mode; configured via hub mcpServers.
- Hub — Host for assistant / block / MCP-server configs.
- Modes (Chat / Edit / Agent / Autocomplete) — Four interaction modes in the IDE extension.
Patterns this full-code implements —
- ★Agent-Computer Interface
IDE-resident Agent mode + standalone `cn` CLI for PR checks.
- ★★Agentic RAG
Continue indexes the codebase using embeddings + keyword search; @Codebase / @Folder triggers similarity retrieval. Custom code-RAG guide documents the indexing pipeline.
- ★★Model Context Protocol
MCP is the canonical custom-tool surface; only available in agent mode.
- ★★ReAct
Agent mode tool-use loop feeds tool outputs back into the model as context.
- ★★Tool Use
Built-in tools + MCP-extensible tools with per-tool permission prompts.
- ★Agent Skills
Continue's rules are the closest primitive to Anthropic Agent Skills: system-message instructions that guide Agent/Chat/Edit behaviour. No verbatim 'skills' equivalent surfaced.
Neighbourhood
Click any neighbour to follow the lineage. Scroll to zoom, drag to pan.