Inception for GitHub Copilot Chat

Use Inception Mercury models directly from the GitHub Copilot Chat model picker in Visual Studio Code.

Setup

1. Install the extension in VS Code 1.125 or newer with GitHub Copilot Chat available.
2. Create an API key in Inception Platform.
3. Open Chat: Manage Language Models, add Inception, name the entry, and enter your API key. Each entry uses only its native credential.
4. Select Mercury 3 in Copilot Chat.

Use Inception: Manage Connection to test the connection, refresh hosted models, and show usage. Inline completions and next edit use the most recently provisioned entry's API key. To use more than one key, add another Inception entry with a different name; the entry name is its identity and VS Code namespaces each entry's models per group.

Group-name-native entries

Entries need only an API key. Rotating a key in Manage Language Models re-scopes the entry's credential automatically, and deleting an entry removes its VS Code-owned credential. No entry IDs, alias blocks, or manual restore steps remain.

Features

• Live chat-only model discovery, with Mercury 3, Mercury 2.5, and Mercury 2 fallbacks when discovery is unavailable.
• Streaming text, reasoning and parallel tool calling, usage reporting, cancellation, and total/idle request timeouts.
• Mercury reasoning choices: Instant, Low, Medium (default), and High.
• Model picker pricing from live discovery (per-million input/output/cached rates) shown in the picker tooltip and fields.
• Merged status bar indicator ($(sparkle) Inception): locally tracked token usage in the text and tooltip, with one menu (Inception: Show Usage) covering usage details, inline-completion toggles, the completion model, and connection actions.
• Local usage tracking across Copilot Chat and inline completions, persisted across sessions, with a dashboard deep link (inceptionCopilot.openUsage).
• Inline autocomplete for any file via the Inception fill-in-the-middle endpoint and Mercury Edit, with debounce, cancellation, and a per-request timeout.
• Next-edit suggestions via the Inception edit endpoint, using recently viewed snippets, a cursor-centered editable region, and your recent edit history.
• Inline-completion controls (toggle either feature, switch the completion model, open settings) live in the merged status bar menu (Inception: Show Usage).
• Optional outcome feedback: accepted suggestions are reported to Inception to improve model quality, sending only outcome metadata (inceptionCopilot.sendFeedback).

Mercury 3 currently advertises a 1,000,000-token context window and 131,072-token maximum output. Mercury 2.5 advertises a 260,000-token context window and 65,536-token maximum output. Mercury 2 advertises 128,000 input tokens and 50,000 output tokens. The default request output budget is 16,384 tokens. Live model limits override fallback metadata. Token counting is an estimate (characters divided by four).

Autocomplete

Inline suggestions come from the Mercury Edit fill-in-the-middle endpoint, separate from Copilot Chat's model picker. Suggestions are insertions at the cursor (or a replace-to-end-of-line while the IntelliSense widget is open); multi-line edits and deletions are not provided.

Next edit

Next-edit suggestions watch your recent edits and cursor positions in other files, send a cursor-centered editable region (default 15 lines) plus recent snippets and a diff history to the Inception edit endpoint, and show the predicted change inline. The stable VS Code inline-completion API can only express insertions and single-line replacements, so multi-line rewrites and deletions are skipped rather than approximated.

Next-edit is off by default — enable it from the status bar menu or with inceptionCopilot.nextEdit.enabled. Autocomplete remains enabled by default and uses the most recently provisioned entry's API key, the same way inline management does.

Using both with Copilot

Multiple inline-completion extensions can compete for Tab. For the best experience, disable other providers (for example editor.inlineSuggest toggles the feature globally, and GitHub Copilot's own suggestions can be turned off per language). Both Inception features can be toggled independently from the status bar menu or disabled in settings.

Usage tracking

The extension tracks tokens and requests locally on this device: every Copilot Chat response plus generated inline autocomplete and next-edit requests accumulate into a per-credential snapshot that survives restarts. The status bar and usage menu show the most recently provisioned entry's credential totals ($(graph) Inception …) with completion feature states in its tooltip, and Inception: Show Usage opens one menu with tracked tokens (input, output, cached, reasoning), an estimated spend, the inline-completion toggles and model picker, and a dashboard deep link (Inception: Open Usage Dashboard opens the Inception Platform usage page). The inceptionCopilot.showUsageStatusBar setting hides the item (the menu stays reachable from the command palette).

Estimates use each model's published rates applied to reported token counts (Mercury 3: $0.10 input, $0.01 cached input, $0.35 output per 1M tokens at the 50%-off launch rate; Mercury 2.5: $0.04 input, $0.004 cached input, $0.15 output at current launch pricing; Mercury 2 and Mercury Edit 2: $0.25 / $0.025 / $0.75). Counts are device-local: they start when the extension first records usage, exclude other tools sharing your key, and do not reflect Inception's billing or the free-token grant — the Inception dashboard is authoritative. No prompts, responses, or API keys are stored; only token counts, model ids, and request outcome metadata.

Settings

Context token counts are estimates. Tool-result text, tool arguments, thinking, and image inputs are included in context budgeting. If the service rejects a full context window, run /compact in Copilot Chat, reduce attachments or tool output, or start a new chat.

Development

Use Node.js 22+ and npm. Run npm ci, npm run check, and npm run package. Press F5 to launch the Extension Development Host. Tests are colocated under src/. See models, development, and security.

API credentials stay in VS Code native secret provider configuration. Requests go directly to https://api.inceptionlabs.ai/v1; no proxy is involved.

This is an independent community extension, unaffiliated with Inception Labs, Microsoft, or GitHub. API usage is billed by Inception.

References

• Inception documentation
• Inception API schema
• Sibling Poolside provider
• Sibling OpenAI OAuth provider

MIT license.