[henrikdukefoss]

Summary

Add an optional Standard Compute memory provider preset using the existing OpenAI Chat Completions transport, following the merged CloudGPU preset in #470 (and Runcrate in #457 before it). Standard Compute is an OpenAI-compatible gateway whose routed StandardCompute model id sends each request to the cheapest capable model.

• memoryProvider: "standardcompute": endpoint default https://api.stdcmpt.com/v1, model default StandardCompute, STANDARDCOMPUTE_API_KEY fallback, explicit env:// key configuration, and a distinct standardcompute session tag.
• Factory, configuration validation, readiness detection, example/template, and all README languages updated.
• Tests mirror the CloudGPU ones: authentication, request serialization (tool_choice: "required"), tool response extraction, overrides, environment-key resolution, and rejection of remote-provider changes from project configuration.
• No defaults change for existing users. Users opt in with their own key (created at https://standardcompute.com/dashboard; docs at https://docs.standardcompute.com); the docs name the destination for conversation data (https://api.stdcmpt.com).

Files changed

• src/services/ai/providers/standardcompute.ts (new): StandardComputeProvider extends OpenAIChatCompletionProvider, overriding endpoint, default model and session tag
• src/services/ai/ai-provider-factory.ts, src/services/ai/provider-config.ts, src/services/ai/session/session-types.ts, src/types/index.ts: wiring and types
• src/config.ts: provider union, STANDARDCOMPUTE_API_KEY fallback, readiness check, config template
• opencode-mem.example.jsonc, README.md, .github/i18n/README.{ar,de,nl,tr,zh}.md: docs
• tests/standardcompute-provider.test.ts (new), tests/ai-provider-config.test.ts, tests/config-resolution.test.ts: tests mirroring the CloudGPU ones

Test plan

• Exercised the change locally: not done. The tests mirror the CloudGPU tests from feat(ai): add CloudGPU memory provider preset #470 but were not run locally; relying on CI. The default model id StandardCompute was confirmed against the public catalogue at GET https://api.stdcmpt.com/v1/models.
• Known limitations: same as other OpenAI-compatible presets; "forceToolChoice": false is documented for models that reject forced tool calls.

Checklist

• Branch is based on main (do not open PRs from commits already on main)
• bun test passes (not run locally; relying on CI)
• bun run typecheck passes (not run locally; relying on CI)
• bun run check passes (not run locally; relying on CI)
• Docs / README updated when behavior or contribution process changes

Disclosure: I work on Standard Compute; this PR was prepared with AI assistance.

🤖 Generated with Claude Code

[Copilot]

🟡 Changes recommended

The key-only Standard Compute configuration leaves runtime provider fields undefined, causing auto-capture, profile learning, and dreaming to skip the provider.

1 open issue

• Resolve Standard Compute preset defaults for runtime callers · New

What changed in this PR

Adds an opt-in Standard Compute memory provider using the existing OpenAI Chat Completions transport, following the CloudGPU and Runcrate presets without changing existing defaults.

Changes:

• Adds provider defaults, factory registration, session tagging, and configuration support.
• Adds tests for authentication, request serialization, response extraction, and configuration handling.
• Documents setup, data transmission, and tool-choice troubleshooting across all README languages.

File	Description
tests/​standardcompute-provider.test.ts	Tests provider behavior and factory registration.
tests/​config-resolution.test.ts	Tests environment-key fallback and project configuration restrictions.
tests/​ai-provider-config.test.ts	Tests minimal configuration and API-key requirements.
src/​types/​index.ts	Extends the provider type union.
src/​services/​ai/​session/​session-types.ts	Adds the Standard Compute session provider type.
src/​services/​ai/​providers/​standardcompute.ts	Implements endpoint, model, and session-tag defaults.
src/​services/​ai/​provider-config.ts	Recognizes Standard Compute as a preset provider.
src/​services/​ai/​ai-provider-factory.ts	Registers and lists the new provider.
src/​config.ts	Adds configuration types, key fallback, readiness handling, and template guidance.
README.md	Documents setup, data destination, and troubleshooting.
opencode-mem.example.jsonc	Adds commented configuration examples.
.github/​i18n/​README.zh.md	Adds Chinese provider documentation.
.github/​i18n/​README.tr.md	Adds Turkish provider documentation.
.github/​i18n/​README.nl.md	Adds Dutch provider documentation.
.github/​i18n/​README.de.md	Adds German provider documentation.
.github/​i18n/​README.ar.md	Adds Arabic provider documentation.

_{Review effort: Balanced}

[EyJunge1]

Fixed in d3b2413: preset endpoint/model defaults are now resolved into runtime CONFIG for all key-only presets (including standardcompute), so auto-capture / dreaming / profile learning see memoryModel + memoryApiUrl. Shared defaults live in memory-provider-presets.ts.

[henrikdukefoss]

On the Copilot finding: it's real, but it isn't specific to this preset. Every key-only preset follows the same path, including cloudgpu (#470), runcrate, cheaper-inference, atlas-cloud and api-route. buildConfig leaves memoryModel / memoryApiUrl undefined, and these guards then skip the external channel:

• auto-capture.ts:715
• user-memory-learning.ts:818
• dreaming-service.ts:163,221
• api-handlers.ts:1721
• user-profile-manager.ts:1601

This PR keeps the existing preset behavior rather than changing it for one provider. The existing config-resolution tests also assert that memoryModel / memoryApiUrl stay undefined for presets.

Happy to do either, whichever you prefer:

1. A separate follow-up PR that resolves preset defaults (endpoint and model) into CONFIG for all presets, with tests that exercise a key-only config through auto-capture.
2. Until then, setting memoryModel and memoryApiUrl explicitly in this preset's README example, so key-only users aren't caught by it.

[EyJunge1]

Pushed a follow-up fix for the Copilot finding: key-only presets now materialize their default memoryModel / memoryApiUrl into runtime CONFIG, so external auto-capture / dreaming / profile learning actually run. Applies to all presets, not only Standard Compute.