* chat: store disk sessions using an 'append log'
This switches from a big blob of JSON to an append-only log of operations
for the chat session. Sending a new request looks something like this:
```
{"kind":2,"k":["requests"],"v":[{"requestId":"request_125e2fa9-1cc2-4197-b023-580eabc9e229","timestamp":1767918567679,"message":{"parts":[{"range":{"start":0,"endExclusive":32},"editorRange":{"startLineNumber":1,"startColumn":1,"endLineNumber":1,"endColumn":33},"text":"what kinds of things can you do?","kind":"text"}],"text":"what kinds of things can you do?"},"agent":{"extensionId":{"value":"GitHub.copilot-chat","_lower":"github.copilot-chat"},"extensionVersion":"0.35.2025120903","publisherDisplayName":"GitHub","extensionPublisherId":"GitHub","extensionDisplayName":"GitHub Copilot Chat","id":"github.copilot.editsAgent","description":"Edit files in your workspace in agent mode","when":"config.chat.agent.enabled","metadata":{"themeIcon":{"id":"tools"},"hasFollowups":false,"supportIssueReporting":true,"additionalWelcomeMessage":{"value":"If handling customer data, [disable telemetry](command:workbench.action.openSettings?%5B%22telemetry.telemetryLevel%22%5D).","isTrusted":{"enabledCommands":["workbench.action.openSettings"]},"supportThemeIcons":false,"supportHtml":false,"supportAlertSyntax":false,"uris":{"command:workbench.action.openSettings?%5B%22telemetry.telemetryLevel%22%5D":{"$mid":1,"path":"workbench.action.openSettings","scheme":"command","query":"[\"telemetry.telemetryLevel\"]"}}}},"name":"agent","fullName":"GitHub Copilot","isDefault":true,"locations":["panel"],"modes":["agent"],"slashCommands":[{"name":"list"}],"disambiguation":[]},"modelId":"copilot/gpt-4o","variableData":{"variables":[{"value":"Repository name: ext-apps\nOwner: modelcontextprotocol\nCurrent branch: main\nDefault branch: main","name":"modelcontextprotocol/ext-apps","modelDescription":"Information about the current repository. You can use this information when you need to calculate diffs or compare changes with the default branch","id":"modelcontextprotocol/ext-apps","kind":"workspace"},{"id":"vscode.prompt.instructions.text","name":"prompt:instructionsList","value":"<instructions>\nHere is a list of instruction files that contain rules for modifying or creating new code.\nThese files are important for ensuring that the code is modified or created correctly.\nPlease make sure to follow the rules specified in these files when working with the codebase.\nIf the file is not already available as attachment, use the #tool:readFile tool to acquire it.\nMake sure to acquire the instructions before making any changes to the code.\n<instruction>\n<description>**CRITICAL**: Read this file FIRST before answering ANY Azure-related questions or performing ANY Azure operations. This file must be loaded as context whenever Azure is explicitly mentioned. It provides accurate and up-to-date information about Azure. **MUST** be read when user mentions: Azure, Azure Functions, AWS Lambda to Azure Functions migration, Azure Static Web App, or any specific Azure service name. Load this instruction file ONLY for Azure-related requests, NOT for generic cloud or deployment questions.</description>\n<file>/Users/connor/.vscode-oss-dev/extensions/ms-azuretools.vscode-azure-github-copilot-1.0.153-darwin-arm64/resources/azureRules/azure.instructions.md</file>\n</instruction>\n</instructions>\n\n","kind":"promptText","modelDescription":"Prompt instructions list","automaticallyAdded":true,"toolReferences":[{"kind":"tool","id":"copilot_readFile","name":"Read File","range":{"start":347,"endExclusive":361}}]}]},"response":[{"kind":"mcpServersStarting","didStartServerIds":[]}],"responseId":"response_8101b2ad-66f7-4e58-ae91-81ff032192e6","modelState":0,"contentReferences":[],"codeCitations":[],"timeSpentWaiting":1767918567679}]}
{"kind":2,"k":["requests",5,"response"],"v":[{"value":"I can help you with a wide range of tasks, including:\n\n1. **Code Assistance**: Writing, debugging, and optimizing code in various languages and frameworks.\n2. **File Management**: Creating, editing, and organizing files in your project.\n3. **Project Setup**: Setting up new projects, configuring tools, and managing dependencies.\n4. **Error Fixing**: Identifying and resolving errors in your code.\n5. **Code Refactoring**: Improving code readability, structure, and performance.\n6. **Documentation**: Generating or improving documentation for your code.\n7. **Version Control**: Assisting with Git operations like commits, branches, and diffs.\n8. **Task Automation**: Writing scripts or setting up tasks to automate repetitive processes.\n9. **Learning and Research**: Explaining concepts, providing examples, or researching solutions.\n10. **Integration**: Connecting APIs, libraries, or services into your project.\n\nLet me know what you need, and I'll get started!","supportThemeIcons":false,"supportHtml":false,"baseUri":{"$mid":1,"path":"/Users/connor/Github/mcp-ext-apps/examples/basic-server-react/","scheme":"file"}}]}
{"kind":1,"k":["requests",5,"result"],"v":{"timings":{"firstProgress":2058,"totalElapsed":3838},"metadata":{"codeBlocks":[],"renderedUserMessage":[{"type":1,"text":"<attachments>\n<attachment id=\"modelcontextprotocol/ext-apps\">\nInformation about the current repository. You can use this information when you need to calculate diffs or compare changes with the default branch:\nRepository name: ext-apps\nOwner: modelcontextprotocol\nCurrent branch: main\nDefault branch: main\n</attachment>\n\n</attachments>\n<context>\nThe current date is January 8, 2026.\nTerminals:\nTerminal: zsh\n\n</context>\n<editorContext>\nThe user's current file is /Users/connor/Github/mcp-ext-apps/examples/basic-server-react/.vscode/mcp.json. \n</editorContext>\n<reminderInstructions>\nYou are an agent - you must keep going until the user's query is completely resolved, before ending your turn and yielding back to the user. ONLY terminate your turn when you are sure that the problem is solved, or you absolutely cannot continue.\nYou take action when possible- the user is expecting YOU to take action and go to work for them. Don't ask unnecessary questions about the details if you can simply DO something useful instead.\nWhen using the insert_edit_into_file tool, avoid repeating existing code, instead use a line comment with \\`...existing code...\\` to represent regions of unchanged code.\n\n</reminderInstructions>\n<userRequest>\nwhat kinds of things can you do?\n</userRequest>\n"},{"type":3,"cacheType":"ephemeral"}],"toolCallRounds":[{"response":"I can help you with a wide range of tasks, including:\n\n1. **Code Assistance**: Writing, debugging, and optimizing code in various languages and frameworks.\n2. **File Management**: Creating, editing, and organizing files in your project.\n3. **Project Setup**: Setting up new projects, configuring tools, and managing dependencies.\n4. **Error Fixing**: Identifying and resolving errors in your code.\n5. **Code Refactoring**: Improving code readability, structure, and performance.\n6. **Documentation**: Generating or improving documentation for your code.\n7. **Version Control**: Assisting with Git operations like commits, branches, and diffs.\n8. **Task Automation**: Writing scripts or setting up tasks to automate repetitive processes.\n9. **Learning and Research**: Explaining concepts, providing examples, or researching solutions.\n10. **Integration**: Connecting APIs, libraries, or services into your project.\n\nLet me know what you need, and I'll get started!","toolCalls":[],"toolInputRetry":0,"id":"e6ff2540-af2b-4fdc-8862-183b6c25c3e9"}],"modelMessageId":"a33d84e3-dbdf-4f75-95dd-bfd96a4cbec2","responseId":"456849bf-a8ec-4149-b76a-b0ddaa1d99ea","sessionId":"53bdd40b-39bd-413d-8069-5562488888a4","agentId":"github.copilot.editsAgent"},"details":"GPT-4o • 0x"}}
{"kind":1,"k":["requests",5,"responseMarkdownInfo"],"v":[]}
{"kind":1,"k":["requests",5,"followups"],"v":[]}
{"kind":1,"k":["requests",5,"modelState"],"v":1}
```
Essentially in this PR there are three parts:
1. Defining the `Adapt` in `chatSessionOperationLog.ts` that defines types
that can translate to/from objects to the on disk state. The diffing
in there was all written by Opus and I'm going to do some more
validation before merging.
2. Making the `requestSchema`. Our models atm are very mutable. This
logic describes how to compare the model objects. It's essentially a
customized deep object equality that we can customize to avoid having
to compare every single property of every object each time (especially
with big tool outputs and images)
This is all type-safe, both for reading the request models and
creating the `ISerializableChatData`.
3. Swapping out to use that log in our logic. The log is a bit stateful
so we pass it around and keep it on the model to re-use next time it's saved.
4. Adding an `appendFile` operation on the IFileService/IFileSystemProvider.
We previously did not have any way to rewrite part of a file. @bpasero
please review 😅
Closes https://github.com/microsoft/vscode/issues/285251
* make it simpler, fix issues
* polish
* address pr comments
* implement IWriteFileOptions.append
* update tests
---------
Co-authored-by: Benjamin Pasero <benjamin.pasero@microsoft.com>
* "Add element to chat" in integrated browser
* Support disabling AI features, PR feedback
* PR feedback
---------
Co-authored-by: Joaquín Ruales <1588988+jruales@users.noreply.github.com>
We have `isDefault` which allows a model to indicate that it is the best/most-recommended model. This PR extends this concept to define default by location, e.g for inline chat a different default model might apply than for panel chat or terminal chat
This makes sure inline chat uses the model the vendor/extension recommends. Also makes sure folks aren't stuck on an old model selection. This can be disabled (via setting) and a custom selection will be honoured for the duration of a vscode session (lifetime of a window)
https://github.com/microsoft/vscode-internalbacklog/issues/6544
- _onDidChangeToolsScheduler.isScheduled is checked to avoid thrashing
setTimeout when disposing many tools
- WorkspaceExtensionsManagementService was listening to file change events
using a debounce. Debounces have overhead because a new timer is
scheduled on every single call. For large amount of file changes
(during EH shutdown when schemas for tools are deregistered) this
caused a notable slowdown. `throttle` should be functionally equivalent.
- avoid triggering input updates (w/ downstream editor effects) each
time the input gets parsed, which happened every time tool is called
- big hammer -- don't bother deregistering MCP tools each time
Results:
- `216ms` to shut down EH before making these changes
- `87ms` in the first three bullets
- `54ms` after skipping MCP tool deregistering. (Basically all the
overhead there was unregistering the JSON schema for tool inputs.)
Added a workspace.azure.yaml Measures marker to telemetry tracking that detects when a workspace contains azure.yaml or azure.yml files. This follows the existing pattern used for other file type detection (like workspace.tsconfig, workspace.npm, etc.).
