[tonydavis629]

Summary

Rebuilt on the embedded provider catalog from #4755, following the contributor guide from #4761:

• pkg/model/provider/catalog/providers.json: an obscurify entry using the existing OpenAI implementation (api_type: openai, base URL https://obscurify.ai/v1, OBSCURIFY_API_KEY, default model deepseek/deepseek-v4-flash, merge_consecutive_messages like the other open-model hosts). It sits after cloudgpu, so it ranks low for automatic selection.
• Tests: a row in openAIAliasProviders (hermetic request test, plus the opt-in live test), rows in the auto-selection tables, a shouldMergeConsecutiveMessages row, and the models.dev exemptions in TestProviderIDsMatchEmbeddedCatalog and modelsDevAbsentProviders. Obscurify.ai isn't in the models.dev snapshot yet, the same situation as runcrate, atlascloud and lunaroute.
• Docs: a provider page, a nav entry, a row in the provider credentials table, the provider lists in docs/configuration/models and the schema description, and an examples/obscurify.yaml recipe.

Obscurify.ai is a privacy-focused provider with an OpenAI-compatible API and 200+ models. Accounts are a random account number with no email, and prompts and responses are not stored on the server.

Testing

On main at 72e52dc, Go 1.27.3:

• go test ./pkg/model/provider/... ./pkg/config/... ./cmd/... ./pkg/runtime/modelcatalog/...: pass. gofmt -l pkg cmd is clean.
• OBSCURIFY_API_KEY=... go test -count=1 -run 'TestOpenAIAliasProvider_LiveAPI$/obscurify' ./pkg/model/provider/: PASS against the real API.
• Built the binary with only OBSCURIFY_API_KEY set, in a clean HOME:
  • docker-agent models lists obscurify deepseek/deepseek-v4-flash as the default.
  • docker-agent run --exec examples/obscurify.yaml asked to read a local file: the agent called read_file and answered with the file's contents.
• Not run: task lint, because golangci-lint isn't installed here. docker-agent run --exec default with only this key still went to DMR when I checked on 7a9206a, and RUNCRATE_API_KEY and GROQ_API_KEY did the same, so I left that alone.

The commit has a DCO sign-off and a verified SSH signature.

Disclosure: I run it.

[aheritier]

👋 Some commits in this PR are not signed and verified by GitHub. Please sign your commits with a GPG or SSH key registered in your GitHub account, then force-push.

Commits that are not verified: 2460796

See GitHub’s guide on signing commits for setup instructions. I’ve added status/needs-signed-commits; it will be removed automatically once every commit in this PR carries a valid GitHub-verified signature.

[aheritier]

👋 This PR has merge conflicts with the base branch. Please rebase or merge the latest base branch and resolve them. I've moved it to draft and added status/needs-rebase; it'll be picked back up automatically once the conflicts are cleared.

[aheritier]

Hi @tonydavis629 we are sorry but we just changed how we manage our providers using a more centralized approach. Could you rebuild our PR with it

Ref: #4755 and #4761

[tonydavis629]

Thanks for the pointer. I rebuilt it on the catalog: Obscurify.ai is now a single entry in providers.json using the OpenAI implementation, with no alias table or client changes. The tests and docs follow the contributor guide. It's one signed and signed-off commit (37e903f) on current main, and the description has the details. Unit tests and the opt-in live test pass. I've marked it ready for review.

[Copilot]

🟡 Changes recommended

The Obscurify live test lacks an explicit opt-in guard, allowing ordinary test runs to send paid API requests whenever OBSCURIFY_API_KEY is exported.

1 open issue

• Require explicit opt-in for Obscurify live API tests · New

What changed in this PR

Adds Obscurify.ai to Docker Agent’s embedded provider catalog, reusing the existing OpenAI-compatible implementation.

Changes:

• Registers the endpoint, API-key variable, default model, and message-merging behavior.
• Extends provider and configuration tests, including models.dev exemptions.
• Adds setup documentation, navigation, and an example configuration.

File	Description
pkg/​model/​provider/​openai/​system_message_merge_test.go	Tests Obscurify message-merging behavior.
pkg/​model/​provider/​openai_alias_providers_test.go	Adds shared request and live-test coverage.
pkg/​model/​provider/​catalog/​providers.json	Registers the provider and its defaults.
pkg/​model/​provider/​catalog_test.go	Exempts Obscurify from models.dev matching.
pkg/​config/​examples_test.go	Allows example models absent from models.dev.
pkg/​config/​auto_test.go	Tests credential discovery and automatic model selection.
examples/​README.md	Lists the new example.
examples/​obscurify.yaml	Provides a tool-enabled agent configuration.
docs/​providers/​overview/​index.md	Lists provider credentials.
docs/​providers/​obscurify/​index.md	Documents setup, usage, models, and limitations.
docs/​data/​nav.yml	Adds provider navigation.
docs/​configuration/​models/​index.md	Adds Obscurify to documented provider lists.
agent-schema.json	Updates the supported-provider description.

_{Review effort: Balanced}

[aheritier]

Reviewed provider dispatch/defaults, credentials and auto-selection, model listing/picker behavior, request/streaming coverage, schema, and documentation against the provider implementation checklist. See inline comments.

CI / operational status: The required gate passed at 37e903f; all executed checks succeeded. The PR is not draft, but GitHub reports mergeable=false and mergeable_state=dirty. Resolve the conflicts and revalidate the resulting head. No local Go tests were run because the available worktree differs from the PR head; review used fetched head sources. No paid inference requests were made.

Issue context: Reviewed #4755 and #4761 and their accessible discussions as implementation/guidance references. No acceptance-criteria issue or Jira reference is linked. Needs-decision: confirm that the PR description is the complete acceptance scope; no additional requirements were inferred.

[aheritier]

[blocking][correctness] Require explicit opt-in before paid live inference. This row leaves liveOptInEnvVar empty, so runOpenAIAliasProviderLiveAPI bypasses its opt-in check and sends a real completion whenever OBSCURIFY_API_KEY is exported. TestOpenAIAliasProvider_LiveAPI runs for every alias, and .env.test does not clear this key, so an ordinary go test ./... or task test can unexpectedly consume the developer's paid balance. Set liveOptInEnvVar: "DOCKER_AGENT_RUN_OBSCURIFY_LIVE_TEST" and require its value to be 1, matching API Route/Lunaroute/OrcaRouter. This also includes Obscurify in the existing credential-alone/no-request regression test. Confidence: high; verified against the row, live-test helper, regression-test selector, and .env.test at this head.

[tonydavis629]

Fixed: the Obscurify row now sets liveOptInEnvVar: "DOCKER_AGENT_RUN_OBSCURIFY_LIVE_TEST". Locally, TestOpenAIAliasProvider_LiveAPI/obscurify skips with only OBSCURIFY_API_KEY exported, and TestOpenAIAliasProvider_LiveAPIRequiresExplicitOptIn/obscurify passes. Thanks for catching it.

[aheritier]

👋 This PR has merge conflicts with the base branch. Please rebase or merge the latest base branch and resolve them. I've moved it to draft and added ; it'll be picked back up automatically once the conflicts are cleared.

[aheritier]

👋 This PR has merge conflicts with the base branch. Please rebase or merge the latest base branch and resolve them. I've moved it to draft and added status/needs-rebase; it'll be picked back up automatically once the conflicts are cleared.

[aheritier]

Reviewed all 13 changed files against the provider implementation checklist, including catalog dispatch/defaults, credentials, selection/discovery, model listing, request/streaming tests, schema, examples, and documentation. No additional demonstrated findings beyond the existing blocking thread.

CI / operational status: Required gate passed at this head; the check rollup is SUCCESS and all executed checks succeeded. Race tests were skipped. The PR is not draft and has no merge conflicts, but GitHub reports mergeable_state=blocked. No local tests were run because the available worktree differs from the PR head. No paid inference requests were made.

Issue context: Reviewed #4755 and #4761 and their accessible discussions as implementation/guidance references; #4756 and #4757 track separate follow-ups. No acceptance-criteria issue or Jira reference is linked. Needs-decision: confirm that the PR description is the complete acceptance scope; no additional requirements were inferred.

Prior threads:

• 4418619721: Still valid but outdated; its root cause is already tracked by current unresolved thread 4432978325, so no duplicate inline comment is added.
• 4432978325: Still open [blocking], confidence high. Reverified at the current head: the Obscurify row omits liveOptInEnvVar, the live-test helper permits inference with credentials alone, and .env.test does not clear OBSCURIFY_API_KEY. The requested explicit opt-in remains necessary.

[tonydavis629]

@aheritier the blocking thread is fixed in ba107af (the Obscurify live test now needs DOCKER_AGENT_RUN_OBSCURIFY_LIVE_TEST=1), and the PR is ready for another look.