VS Code Adds BYOK for Copilot and Streaming

Visual Studio Code 1.117, released April 22, 2026, adds three targeted improvements: Bring Your Own Key (BYOK) for Copilot Business and Copilot Enterprise, experimental incremental chat rendering, and the ability to launch the Copilot CLI from any terminal profile. The BYOK feature lets teams plug their own model API keys directly into VS Code chat, supporting providers such as OpenRouter, Ollama, OpenAI, and Google. Incremental rendering streams chat responses block-by-block to reduce perceived wait time. The terminal change fixes a long-standing startup failure when non-default shells like fish or Git Bash are used.

BYOK is enabled by default and is controllable via the administrative policy named "Bring Your Own Language Model Key" in the organization settings on GitHub.com. After admin enablement, members can configure or install language model provider extensions and supply provider API keys inside VS Code, routing chat requests to those external endpoints rather than Microsoft-hosted model endpoints. This supports both cloud-hosted providers and on-prem or local inference stacks like Ollama.

Incremental response rendering is exposed as experimental settings. Key config options are chat.experimental.incrementalRendering.enabled (default: true), chat.experimental.incrementalRendering.animationStyle (options: none, fade, rise, blur, scale, slide, reveal; default: fade), and chat.experimental.incrementalRendering.buffering (options: off, word, paragraph; default: off). The implementation renders content block-by-block as tokens arrive instead of relying on timer-based flushes, so longer outputs appear sooner and feel more conversational.

The Copilot CLI can now be launched from any terminal profile in VS Code, addressing a bug where non-default shells prevented the terminal-based CLI from starting. This is a UX and developer workflow improvement rather than a change to the CLI itself.

The BYOK capability is a concrete step in the broader industry trend toward developer and enterprise control over model choice and data locality. By letting organizations use their own keys and model providers, Microsoft reduces backend load and gives teams options for compliance, cost optimization, and latency improvements. Integration with local inference tools like Ollama lets developers run open-source models on-prem or on-device, which matters for sensitive code and regulated industries.

Incremental rendering is a user-facing optimization that mirrors improvements seen across chat UIs and LLM clients; practitioners get a smoother developer experience without changing model architectures. The configurable buffering and animation settings let teams balance immediacy against the risk of displaying partial or malformed Markdown.

Administrators should audit and decide whether to enable BYOK and which providers to allow via policy, since BYOK changes where inference occurs. Watch for extension ecosystem updates that add provider integrations, and monitor adoption of local inference stacks like Ollama inside enterprise developer environments. Security and compliance teams will need to update guidance on secret management and API key rotation when keys are stored or referenced from developer machines.

This is a pragmatic release: 1.117 prioritizes operational flexibility and incremental UX gains over new model capabilities. For teams that need model choice, on-prem inference, or better Copilot streaming behavior, the update materially improves daily developer workflows.