AgentLink
A coding-agent harness built into VS Code, with standalone Desktop and CLI previews.
Run frontier coding agents with the editor's intelligence, visible execution, reviewable changes, and as much, or as little, supervision as the task needs. Use the macOS Desktop app for standalone Ask Agent chat or the Apple Silicon CLI for terminal coding when you do not need an editor.
Get started · Documentation · Why AgentLink · Releases
Why AgentLink
A capable model is only part of a capable coding agent. It also needs a good working environment: editor intelligence instead of text guesses, fast feedback instead of a surprise broken build, and a way for you to supervise or redirect work without taking the controls away.
AgentLink works through VS Code. It gives an agent language-server navigation and diagnostics, opens edits in native diff views, runs commands in the terminal you can see, and keeps activity, decisions, and recovery close to the work.
That is the product bet: better context, better feedback, and better control make agents more useful on real codebases—not just more autonomous in a demo. Read the concise case for AgentLink.
What makes it different
The editor is the runtime
Give the agent the same semantic understanding you use: definitions, references, symbols, type information, code actions, diagnostics, and workspace-aware rename. Proposed edits arrive as diffs you can accept, reject, or adjust yourself.
Autonomy is a dial, not a switch
Review every action when the task is risky, or let familiar work move faster with focused rules and Approve for Me. Commands remain visible, approvals carry your feedback back to the agent, and checkpoints make experimentation reversible.
Nothing happens in the dark
Tool calls, progress, questions, approvals, queued work, and background agents stay visible in the chat and Activity Shelf. Use the browser remote to check a session, answer a question, or inspect a read-only diff without taking over the editor.
A second opinion from another model provider
Use OpenAI/Codex alongside models from configured OpenAI-compatible providers or external ACP agents. AgentLink can route review to another provider, giving important work an independent set of model blind spots.
Context for the code, not the harness
AgentLink keeps local lexical and structural codebase retrieval on your machine, progressively discloses tools, bounds noisy terminal output, and condenses long sessions without losing the active task.
What you can do
| Work | AgentLink gives you |
|---|
| Build in the editor | VS Code chat, language intelligence, diagnostics, diff review, integrated terminals, and modes for coding, planning, debugging, review, and questions. |
| Stay in control | Inline approvals, editable command requests, audit context, checkpoints/revert, structured questions, and visible task status. |
| Work in parallel | Background research and review, cross-provider review, Fleet progress, steering, and bounded results. |
| Connect your workflow | MCP tools, resources, and prompts; AGENTS.md/CLAUDE.md; skills, custom modes, slash commands, hooks, and Agent Plugins. |
| Use your accounts | ChatGPT/Codex, OpenAI, and compatible providers on your own accounts. Local retrieval and telemetry stay local by default. |
Get started
Choose the surface that fits your work: the VS Code extension has the fullest editor and review integration; Desktop offers Ask Agent chat and a view into running VS Code sessions; the CLI runs a local coding session in your terminal. Desktop and CLI are separate, narrower macOS previews.
Install the VS Code extension
The installer selects the target for the VS Code extension host on this machine:
curl -fsSL https://raw.githubusercontent.com/reefbarman/agentlink/main/scripts/install.sh | bash
For a remote or emulated extension host, set its target explicitly:
curl -fsSL https://raw.githubusercontent.com/reefbarman/agentlink/main/scripts/install.sh \
| AGENTLINK_VSCE_TARGET=linux-x64 bash
The installer verifies the release checksum when one is published. Add --version X.Y.Z to pin a release or --dry-run to preview what it would install. You can also download a matching .vsix from the latest release:
code --install-extension agentlink-*.vsix --force
This is the beginning of the README. Read the rest at the original. Read the full README ↗
