Skip to content

WSO2 Integrator Copilot Features

This page is a reference for the user-facing features of the Copilot panel. For how to use the three modes, see Modes.

Checkpoints and undo

The Copilot creates two kinds of checkpoints so you can roll back changes at different granularities — a fine-grained one for each turn the Copilot edits files, and a coarse-grained one for each message you send.

Per-turn checkpoint — artifact changes

Every time the Copilot adds, edits, or deletes an artifact in your project, it snapshots the previous state of the affected files. When the turn finishes, a Checkpoint card appears in the conversation listing the files that changed and how many lines were added or removed.

Per-turn checkpoint card

The card has two actions:

  • Accept — acknowledges the changes and dismisses the card. Files stay as they are.
  • Undo — rolls back every file the Copilot changed in that turn. Works across multiple tool calls, file creations, edits, and deletions made in the same turn.

Per-turn checkpoints cover any project mutation: file writes, file edits, new data mappers, connector additions, and connector removals. If you edit the same files yourself between the turn and hitting Undo, the Copilot detects the conflict and asks whether you want to force the rollback or keep your manual edits.

Per-message checkpoint — rewind project and chat

Before each message you send, the panel snapshots both your project state and the chat history. This lets you rewind an entire multi-turn exchange back to the moment before a specific message — not just the most recent edit.

Per-message checkpoint

Click the rewind icon next to an earlier user message to:

  • Revert every file in your project to the state it was in when you sent that message.
  • Trim the chat history so the conversation returns to that point. Messages and replies sent after it are removed.

Reach for this when a series of follow-ups has gone down the wrong path and you want a clean reset before re-prompting, rather than undoing one turn at a time.

Info

Both kinds of checkpoints are stored on disk and survive VS Code restarts. The Copilot keeps the most recent checkpoints and garbage-collects older ones automatically.

Plan approval

When the Copilot finishes writing a plan in Plan mode, it presents a Plan Approval dialog.

Plan approval dialog

Your options:

  • Approve Plan — confirms the plan. The Copilot exits Plan mode, and you typically switch to Edit mode to execute it.
  • Request Changes — opens a feedback box. Type what you want changed (for example, "Don't touch the existing unit tests; I'll update them separately"), and the Copilot revises the plan and re-asks for approval.

The Copilot may also ask for permission to enter or exit Plan mode if it thinks the task warrants a switch. These use the same approval dialog with a different prompt.

Todo tracking

For multi-step tasks, the Copilot maintains a todo list visible in the conversation. Each item shows its status — pending, in progress, or done — as the Copilot works through them. This gives you a live view of what's happening during longer operations.

Todo list

Todos are in-memory for the current turn — they're a progress indicator, not a persistent task tracker.

Attachments

You can attach files and images to a message by clicking the Attach button in the input bar.

Attachments

Supported types

  • Text files — .txt, .md, .csv, .html, .js, .ts, .css, .json, .xml, .yaml, and 12 other common text formats. The Copilot reads the contents as part of your message.
  • PDF — parsed as a document (vision-based).
  • Images — .png, .jpg, and similar. The Copilot uses vision to read diagrams, screenshots, UI mockups, and documents-as-images.

Attachments are validated before the message is sent — if anything is unsupported or malformed, the Copilot tells you upfront instead of partially failing halfway through a generation.

Typical uses

  • OpenAPI specs and JSON/XML schemas for generating REST APIs and transformations
  • Screenshots of error dialogs when asking the Copilot to help debug
  • Architecture diagrams when asking the Copilot to scaffold a project from a design
  • Sample request/response payloads when building mediations

@Mentions

Type @ in the input bar and start typing a file or folder name to mention it inline. Mentions are picked from your open project.

File and folder @mentions

Mentions are a quick way to point the Copilot at a specific file or directory without having to describe it. For example:

Refactor @src/main/wso2mi/artifacts/sequences/TransformOrders.xml to use the data mapper.

The Copilot reads the mentioned file automatically — you don't need to attach it.

Project instructions with AGENTS.md

You can pin project-specific instructions for the Copilot by creating an AGENTS.md file at the root of your project. This follows the same convention used by other AI coding tools (such as Claude Code's CLAUDE.md and the AGENTS.md spec), so a single file can guide multiple tools.

Use it to capture the conventions you'd otherwise have to repeat in every prompt — for example:

  • House style and naming conventions ("Name all sequences in PascalCase and suffix them with Seq.")
  • Preferred approaches ("Always use the data mapper for JSON-to-XML transformations rather than PayloadFactory.")
  • Project-specific context ("Connections live in src/main/wso2mi/artifacts/local-entries; reuse OpenWeather instead of creating new ones.")

The Copilot treats AGENTS.md as authoritative guidance layered on top of its built-in system prompt: on conflict, your instructions take precedence.

How it's loaded

  • AGENTS.md is loaded automatically — there's nothing to attach or @mention. Just keep the file at the project root.
  • It is re-sent to the Copilot only when its contents change, so it doesn't consume tokens on every turn.
  • Edit the file and the Copilot picks up the new instructions on your next message. Delete it and the Copilot discards the project-level instructions it had sourced from it.

Size limit

AGENTS.md is capped at 30 KB. If the file is larger, the Copilot uses only the first 30 KB and shows an AGENTS.md truncated notice in the conversation so both you and the Copilot know that rules beyond that point are not in context. Keep the file focused and concise.

Sessions

Each conversation you have with the Copilot is a session. Sessions are saved automatically and you can switch between them from the session switcher in the header.

Session switcher

  • New Chat (in the header) starts a fresh session. Your current session is saved to history.
  • The session switcher groups history into Today, Yesterday, Past Week, and Older.
  • Session titles are derived from your first message (or a short fallback label). You can switch between sessions freely; the Copilot remembers the full history per session.

Note

Sessions are append-only — you can't edit or delete individual messages. If a conversation has gone in the wrong direction, start a new session.

Adaptive thinking

For complex requests, the Copilot can use adaptive thinking — it works through the problem before responding, and skips the step when the request is straightforward. When it does think, a collapsible Thinking segment appears in the conversation with the Copilot's reasoning. Expand it to see the approach, or collapse it to focus on the final answer and the changes.

Thinking segment in the conversation

Adaptive thinking is on by default. You can toggle it from the Settings gear in the panel header. Turn it off if you find the Copilot is overthinking — for example, deliberating on small edits where you'd rather it just make the change.

Adaptive thinking toggle in settings

Shell command approvals

In Edit mode, the Copilot can run shell commands — typically build, test, or project-setup commands. Commands are classified into three tiers:

  • Auto-approved — standard, read-only, or project-scoped commands (for example, ls, mvn clean install, pnpm test). These run without interruption.
  • Needs approval — commands that write outside the project, use the network, or look unfamiliar. The Copilot pauses and asks for your approval, showing the exact command. You can approve once, approve for the rest of the session, or deny.
  • Blocked — anything genuinely dangerous: elevated commands (sudo), interactive editors (vim, nano), access to sensitive paths (~/.ssh, ~/.aws, .env), or commands that try to escape the project directory via symlinks. These are refused outright and the Copilot explains why.

In Ask and Plan modes, only read-only shell commands are permitted.

Model selection

Open Settings from the header and find the Models section to choose which Claude model the Copilot uses.

Model settings

Two presets are configurable:

  • Main model — used for the primary conversation. Options include Claude Sonnet (default, best balance of speed and quality) and Claude Opus (stronger reasoning, slower).
  • Sub-agent model — used when the Copilot spawns a sub-agent for codebase exploration or deep documentation lookups. Options include Claude Haiku (default, fast) and Claude Sonnet (more thorough).

You can also override with a custom model ID if your account supports a specific model. Your selection is saved locally per device.

Usage limits and quota

The behaviour here depends on how you signed in — see Sign in for the three options.

WSO2 Account

  • Your account includes a free usage quota that resets on a recurring schedule.
  • The Copilot shows your remaining quota and reset time when you open Settings.
  • When you run out, the panel displays a Usage limit reached message explaining when your quota resets. You have three options:
    • Wait — your quota refreshes automatically at the displayed reset time.
    • Switch to Anthropic API key — unlimited, billed to your Anthropic account.
    • Switch to AWS Bedrock — unlimited, billed to your AWS account.

Anthropic API key

  • No WSO2-side quota. Usage is billed directly by Anthropic based on your key's plan.
  • If your key is revoked or rate-limited, the Copilot signs you out and asks for a new key.

AWS Bedrock

  • No WSO2-side quota. Usage is billed through your AWS account.
  • If your credentials expire (for example, a session token runs out), the Copilot signs you out and asks for new credentials.

AI usage and data handling

For details on how your prompts and attached files are processed, see AI Usage and Data Handling Guidelines.

Long conversations

The Copilot handles very long conversations automatically. When a session gets close to the context limit, it summarizes older messages in-place without interrupting you — you'll see a Compacted earlier messages segment in the conversation where the summary was written. The Copilot keeps going from there with the summarized history plus your newer messages.

You don't need to do anything; this is entirely automatic. If you prefer a fresh slate at any point, click New Chat in the header.

Reconnecting after closing the panel

If you close the Copilot panel or VS Code while the Copilot is working on a long task, you can reopen the panel and the Copilot will reconnect to the in-flight run. The conversation picks up where you left off, including streaming events and tool calls. If the task already finished while the panel was closed, you'll see the final results when you reopen.

What's next