You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
AbliteratedLLM (https://abliteratedllm.dev) is a pay-per-token, OpenAI- and Anthropic-compatible API for abliterated (refusal-removed) and stock open-weight models, prepaid in USD by card or crypto.
This adds AbliteratedLLM as an ordinary API-key provider. It shows up in ocx init, in ocx provider add abliteratedllm, in the dashboard's Add provider picker and in API-key login. The preset reuses openai-chat at https://api.abliteratedllm.dev/v1. It adds no sponsor status, ranking preference, OAuth flow, adapter or wire pin. It follows the shape of cheapaimodel (#7899), Modelflare (#7037), AI Stupid Level (#7623) and TokenLab (#6221), and handles key validation the way OpenGateway does, because the model list is public.
Chat Completions only. Every inbound (Responses, Chat, Anthropic) stays on POST /v1/chat/completions, which serves every catalog model. That is the same wire the other OpenAI-compatible gateway presets use. The service also has POST /v1/responses and an Anthropic POST /v1/messages, but the preset doesn't use them. The Responses endpoint is stateless (store is forced to false and previous_response_id is rejected). A test pins Chat for all 15 models and all three inbounds.
Discovery. Live discovery is off (liveModels: false, owner decision 2026-11-18); the preset uses the seeded list below. GET /v1/models is public: on 2026-11-14 it returned 200 with no key and also with an invalid Bearer value. It can't verify a key, so the entry sets apiKeyValidation: "unknown", the same as OpenGateway. validateApiKey then returns unknown without making a request, so the login flow reports the key as unverifiable instead of giving a false positive. The service does have a key-gated GET /v1/balance, which answers 401 for a missing or invalid key. The registry has no per-entry validation URL, though, and I didn't add one in this PR (see the end of this description).
Seeded models. The full catalog from GET /v1/models on 2026-11-14, in catalog order: glm-5.3-flash-abliterated (the default), deepseek-v4.1-flash-1017, deepseek-v4.1-flash-1017-ultrafast, deepseek-v4.1-flash-original, glm-5.3-flash, glm-5.3, glm-5.3-fast, kimi-k3, kimi-k3.1-pro, kimi-k3-fast, ember-1, muse-spark-open, qwen3.8-max, qwen3.8-2.4t-a95b, minimax-m3. This follows feat(providers): add cheapaimodel API-key preset #7899, which seeded its whole live catalog. The preset keeps this seeded list; live discovery is off. glm-5.3-flash-abliterated is the service's abliterated model; the other 14 are the vendors' original weights.
Declared limits. Context windows come from each row's context_length: 1,048,576 tokens, except 262,144 for the two Qwen3.8 models and 512,000 for minimax-m3. Only glm-5.3-flash-abliterated gets an output cap (131,072). Every other row lists max_output_tokens equal to its window, so no cap is declared for them. Input modalities come from input_modalities. glm-5.3 and glm-5.3-fast are text only, so images for them go to the vision sidecar. The gateway would answer an image part sent to them with 400 unsupported_content. glm-5.3-flash also lists video, but it is seeded as text and image because the Codex catalog only carries text, image and audio. The seeds live in model-seeds.ts, as the Opper, AI Stupid Level and Morph catalogs do.
Reasoning. The Chat Completions reference documents reasoning_effort as low, medium or high for every model (some models also take none). The preset therefore sets reasoningEfforts: ["low", "medium", "high"], which gives:
xhigh and max are sent as high
minimal is sent as low
none omits the field
Custom destinations are kept.preserveCustomDestination leaves an existing custom provider with the same name on its own URL.
Routing statement. The entry note and the guide paragraph say what GET /v1/models showed on 2026-11-15. Of 15 models, glm-5.3-flash-abliterated is the only abliterated one (variant: "abliterated"), and the other 14 have provider_label: "Stock model". None was labelled "Official DeepSeek API", the privacy policy's label for requests forwarded to DeepSeek's own API. The listed models, the DeepSeek ones included, run on Fireworks AI with zero data retention. Nothing is promised for models the service lists later. A test pins these sentences.
This revision follows the owner's decision of 2026-11-15: a canonical preset, with the legal-entity evidence item waived. The free-directory reference row from the previous revision is dropped, so the service is listed once. The branch holds the two reviewed commits (4bf2b3473, 90c6e3f3a), rebased on dev04304da28 on 2026-11-18. The guide text now lives on the API-key catalog page from #8768.
Files:
File
Change
src/providers/registry/entries-extended.ts
the abliteratedllm entry, after cheapaimodel, with the routing note
src/providers/registry/model-seeds.ts
the seeded model list, context windows and input modalities
tests/providers/abliteratedllm-provider.test.ts
new test covering: entry points, seeds, windows, output cap and modalities; the Chat wire for all 15 models and inbounds; the reasoning ladder on the built wire; custom destination; the seeded list without a catalog fetch; unknown key validation without a probe; the routing sentences in the note
the other 7 guides/providers.md, all 9 getting-started/quickstart.md, structure/ops/docs-and-release.md
preset counts recounted from PROVIDER_REGISTRY on current dev: 109 total and 88 key-based (dev: 108 and 87)
Evidence
Listed in the order MAINTAINERS.md gives. The endpoints were checked on 2026-11-14 and the routing on 2026-11-15.
Documented OpenAI-compatible endpoints.
POST /v1/chat/completions: https://api.abliteratedllm.dev/v1/chat/completions with a Bearer key. Streaming and function tools are supported.
GET /v1/models, not used for discovery (liveModels: false). It is public: it returned 200 without a key and with an invalid key on 2026-11-14. This is why the entry declares apiKeyValidation: "unknown" and not the authenticated discovery that the contributing guide describes.
GET /v1/balance is key-gated: it returned 401 invalid_api_key without a key and with an invalid key on 2026-11-14.
Terms of service and operating legal entity. The Terms of service, §1, say the service is operated by its founder while AbliteratedLLM LLC, a proposed Wyoming company, is being formed. The owner waived this item for AbliteratedLLM on 2026-11-15. I'll cite the registration here once the LLC exists.
Resale or routing authorization. AbliteratedLLM serves open-weight models. Per its privacy policy, the abliterated model runs on dedicated GPUs and the stock models on shared GPUs, both at Fireworks AI with zero data retention. Routing as observed on 2026-11-15 is described above.
Verification date. 2026-11-14 for the contract and 2026-11-15 for the routing, recorded in the code comments on the entry and the seed list.
Verification
Run with Bun 1.4.4 on Windows, after bun install --frozen-lockfile (and in gui/, as CI does), first on dev1d10758fa. After the rebase onto 04304da28 (0 behind), I re-ran the first row (176 pass), the typecheck (no type errors in 2,566 files), structure:check and privacy:scan, and all passed.
Command
Result
bun test over abliteratedllm-provider, provider-registry-parity, free-directory-runcrate, docs-provider-preset-counts, docs-spanish-localization, file-size-ratchet, test-layout and test-layout-tooling
176 pass, 0 fail
bun test over all 65 tests/providers and tests/ci-workflows files that import the registry, plus the free-directory tests
1534 pass, 0 fail
bun run typecheck
no type errors in 2,533 files (2,566 after the rebase)
bun run privacy:scan, structure:check, skill:surface:check, docs:clients:check
all passed
I didn't run the full suite (bun run test), the changed-tests lane or the docs-site build on this revision; they are left to CI. On the earlier revision the docs-site build passed: 766 pages, 115,625 internal links. No live end-to-end test with a real AbliteratedLLM key was run. This change was prepared by an AI coding agent on my behalf.
Maintainer update (2026-11-18)
Owner decision: the routing-authorization evidence item is waived, along with the legal-entity item waived on 2026-11-15. The preset uses its seeded model list (liveModels: false), because the public GET /v1/models cannot validate a key. Maintainers pushed 6165691, which sets liveModels: false, updates the note and the en/es catalog paragraphs, and replaces the live-catalog test with a seeded-list test, and 0c3ba32, which merges dev, keeps both the AbliteratedLLM and LLMTR layout registrations, and updates the seed comment.
Checklist
Scope stays focused and avoids unrelated cleanup.
Docs or release notes were updated when needed.
Security-sensitive changes were reviewed for secrets, auth, and unsafe defaults (no secrets; the user's key is sent only to https://api.abliteratedllm.dev/v1, and key validation never sends it anywhere because it reports unknown without a request).
Review readiness checklist
This PR stays in draft until every box below is ticked. Tick all four boxes once the requirements are met:
Required local validation passed; commands, results, and any full-suite exception are documented.
I pushed my PR to a recent dev commit (at most 10 behind; a maintainer may still ask for the exact tip before merge).
I resolved all correct Codex and CodeRabbit findings.
My PR is ready for review.
Readiness boxes ticked by maintainer @lidge-jun at integration (2026-11-18) after the exact-head review and CI described in the PR comments.
It looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the reviews.auto_review.auto_pause_after_reviewed_commits setting.
No actionable comments were generated in the recent review.
Walkthrough
The registry adds AbliteratedLLM with seeded models, live model discovery, and Chat Completions routing. Provider tests cover its configuration and behavior. Documentation updates preset totals and describes the provider.
Changes: AbliteratedLLM provider
Layer / File(s)
Summary
Registry entry and behavior tests src/providers/registry/model-seeds.ts, src/providers/registry/entries-extended.ts, tests/providers/abliteratedllm-provider.test.ts, tests/fixtures/test-layout-expected.json, scripts/test-layout/layout.json
The registry adds 15 seeded models with context windows and input modalities. It registers AbliteratedLLM with Chat Completions routing, live model discovery, custom-destination preservation, and unknown key-validation status. Tests check provider metadata, model routing, reasoning-effort mapping, catalog fetching, and validation behavior.
Provider documentation and preset counts docs-site/src/content/docs/guides/api-key-catalog.md, docs-site/src/content/docs/getting-started/quickstart.md, docs-site/src/content/docs/*/getting-started/quickstart.md, docs-site/src/content/docs/*/guides/providers.md, structure/ops/docs-and-release.md
The API-key catalog documents AbliteratedLLM’s setup, endpoint, model catalog, and validation behavior. Quickstarts and provider guides update the total from 108 to 109 and the key-based count from 87 to 88. The other documented category counts remain unchanged.
sequenceDiagram
participant OpenCode client
participant Provider registry
participant AbliteratedLLM API
Provider registry->>AbliteratedLLM API: GET /v1/models for live discovery
OpenCode client->>Provider registry: Route request for a model
Provider registry->>AbliteratedLLM API: Send Chat Completions request with upstream model ID
Loading
Merge Risk: Minimal · previous review 53dee
This adds a new AbliteratedLLM API-key provider preset and updates the documentation counts. No concrete merge-blocking risk was found, and the change is additive and easy to revert.
Pre-merge checks: 4
Check name
Status
Explanation
Title check
The title "feat(providers): add AbliteratedLLM API-key preset" matches the main change. The PR adds AbliteratedLLM as a canonical API-key provider using openai-chat, with a registry entry, model seeds…
Linked Issues check
Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check
Check skipped because no linked issues were found for this pull request.
Description Check
Check skipped - CodeRabbit’s high-level summary is enabled.
Full details: Title check
Explanation
The title "feat(providers): add AbliteratedLLM API-key preset" matches the main change. The PR adds AbliteratedLLM as a canonical API-key provider using openai-chat, with a registry entry, model seeds, docs, and tests. The title is short, specific, and states the primary change. The docs count updates (108 to 109 presets) and the test layout change are supporting edits that the title does not need to list.
Finishing Touches
Generate unit tests (beta)
Create a new PR
Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.
Reviewing files that changed from the base of the PR and between 174f22d and 53dee36.
Files selected for processing: 1
tests/fixtures/test-layout-expected.json
Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 9 remain after this review.
450-450: Minor · Functional Correctness · Quick win
Unverified finding
Verification ran but could not confirm this finding. It is shown for review, not as a verified issue.
Clarify that key validation is not available for every preset.
This line says the picker validates each credential. AbliteratedLLM uses apiKeyValidation: "unknown", and Line 1029 says its public model endpoint cannot validate a key. Users may believe an invalid key was verified and only discover the problem during inference. State that validation depends on the provider and may be unavailable.
As per coding guidelines: “Document current shipped or intentionally pending behavior.”
Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 9 remain after this review.
Merge-manager review: the owner accepted AbliteratedLLM as a canonical preset in the 2026-11-14/15 backlog review, conditional on the MAINTAINERS.md evidence check. The code is ready (entry, 15 seeds matching the live catalog, layout registration, tests). Evidence status:
Endpoints and /v1/models: present (public listing disclosed, apiKeyValidation: "unknown").
Maintenance owner and verification date: present.
Operating legal entity: missing. Terms §1 says the service is operated by its founder and "AbliteratedLLM LLC is a proposed limited liability company… being formed".
Routing statement needs a correction: the privacy policy says models labelled "Official DeepSeek API" go to DeepSeek (PRC storage, possible training use), which contradicts "every model runs under zero retention". Please check each seeded id's provider_label in /v1/models and say in the entry note and guide which seeds, if any, route to the official DeepSeek API.
The entity item is a hard requirement for a canonical row, so this is held until the LLC exists (cite its registration) — or, as you offered, we can take it now as an inert src/providers/free-directory.ts reference row and promote it later. Let me know which you prefer.
Two merge mechanics apply to every preset PR right now: docs-site/src/content/docs/guides/providers.md is 1997 lines on dev and the file-size ratchet fails it at 2000, so a new preset's paragraph must stay within the remaining room or go to its own docs page; and preset counts in the quickstarts/guides must be recounted from the registry at merge time (tests/ci-workflows/docs-provider-preset-counts.test.ts), since parallel preset PRs each write "dev + 1".
whiteworthy
changed the title
feat(providers): add AbliteratedLLM API-key preset
feat(providers): list AbliteratedLLM as a free-directory reference row
Nov 15, 2026
Thanks for the review. Let's take it now as the inert reference row. I've rebuilt the branch on dev as one commit (3d7bf57):
The row:abliteratedllm in free-directory.ts, group signup-credit, reference defaults, no seeds and no registry entry.
A test: pins the row's fields and checks that abliteratedllm is not in PROVIDER_REGISTRY.
No docs changes: the guide limit and the preset counts don't come into it.
The description is updated.
Routing. None of the listed models goes to the official DeepSeek API. On 2026-11-15, /v1/models lists 15 models. glm-5.3-flash-abliterated is the only abliterated one, and the other 14 have provider_label: "Stock model"; none is labelled "Official DeepSeek API". The DeepSeek models run on Fireworks serverless inference with zero data retention, like the other stock models. The "Official DeepSeek API" label in the privacy policy is for models routed to DeepSeek's own API, and none is listed. The row's note says this, with the date, and the code comment cites the policy and /v1/models.
Entity. I'll open a follow-up to promote the row to a preset once the LLC is registered, and cite the registration there.
Update from the merge manager: #8768 (merged as bd712ac) moved the API-key preset catalog out of guides/providers.md into docs-site/src/content/docs/guides/api-key-catalog.md (Spanish: es/guides/api-key-catalog.md), so the 2000-line limit no longer constrains a new preset paragraph. When you next merge dev, put the preset's section on that page; the preset counts are now checked against that page by tests/ci-workflows/docs-provider-preset-counts.test.ts, so recount from the registry at that point. The evidence items in my earlier comment still apply.
Thanks for checking the live catalog — that settles the routing question: on 2026-11-15 none of the 15 seeds carried "Official DeepSeek API", so the corrected wording (listed models run on Fireworks AI with zero retention; no promise beyond what was observed) is right.
Owner decision (lidge-jun, 2026-11-15): the owner waives the legal-entity evidence item for AbliteratedLLM and wants it as a canonical preset, not a free-directory row. Could you restore the registry entry, seeds, docs and tests from your reviewed head d176b49ccf (that code was ready), carry the corrected routing note into the entry note and the guide paragraph, and drop the free-directory row so the service isn't listed twice? The guide paragraph now goes in docs-site/src/content/docs/guides/api-key-catalog.md (+ Spanish es/guides/api-key-catalog.md) after #8768, and the preset counts should be recounted from the registry against current dev. Then tick the readiness boxes on that head; I'll record the waiver in the merge comment and integrate on green CI.
whiteworthy
changed the title
feat(providers): list AbliteratedLLM as a free-directory reference row
feat(providers): add AbliteratedLLM API-key preset
Nov 15, 2026
Thank you for the waiver. I've done the rework on dev1d10758fa:
Branch: rebuilt as the two reviewed commits, 2d4edb28f and 174f22d55. The free-directory row is gone.
Routing note: the corrected statement is in the entry note, in the guide paragraph on guides/api-key-catalog.md and in the Spanish catalog page, and a test pins it.
Counts: recounted from the registry, to 109 total and 88 key-based.
The targeted tests (176 pass), the wider provider and ci-workflow set (1534 pass), the typecheck, and privacy:scan, structure:check, skill:surface:check and docs:clients:check all pass locally. The full suite and the docs-site build are left to CI. I've ticked the readiness boxes and marked the PR ready.
Owner decision (2026-11-18), on top of the earlier legal-entity waiver: the routing-authorization evidence item is waived, and the preset keeps a seeded model list. The public GET /v1/models answers without a key, so it cannot satisfy the authenticated-discovery rule, and ocx provider test would report connected with a bad key. I pushed 6165691 on top of a dev merge. It sets liveModels: false, updates the entry note and the English and Spanish catalog paragraphs, and replaces the live-catalog test with one that checks the 15 seeded models are listed without a fetch. Locally the preset, registry parity, preset-count, Spanish localization, ratchet and layout tests, typecheck and privacy:scan pass. I ticked the readiness boxes at integration. This merges after exact-head CI and the review round.
Add AbliteratedLLM as an ordinary API-key preset on openai-chat at
https://api.abliteratedllm.dev/v1. Discovery reads the public GET /v1/models,
which answers 200 for any Bearer value, so key validation reports unknown.
Seed the 15-model catalog listed on 2026-11-14 with its context windows,
input modalities and the default model's output cap, and pin the documented
low/medium/high reasoning_effort ladder.
The preset note and the catalog entry state the routing observed on
2026-11-15 instead of promising zero retention for every model:
GET /v1/models listed glm-5.3-flash-abliterated (variant "abliterated") and
14 models with provider_label "Stock model". None was labelled "Official
DeepSeek API", the privacy policy's label for requests forwarded to
DeepSeek's own API; the listed models, the DeepSeek ones included, run on
Fireworks AI with zero data retention.
The English and Spanish entries go on guides/api-key-catalog.md. Every
locale's provider guide and quickstart, and structure/ops/docs-and-release.md,
recount the documented totals to 109 presets and 88 key-based.
Maintainer integration record (lane L3, 2026-11-19).
Integrating exact head 61ccae5a8a5744855ac31914140202e1ac2776c4 under the maintainer-integration path, following the owner decision: routing authorization waived, with liveModels: false and seeded models only.
Review. The same gpt-6.2-sol reviewer covered every round. The merge check at this head gave APPROVE: one Spanish catalog conflict was resolved, and counts match the merged registry (111 total, 89 key-based, 18 OAuth).
Behavior.ocx provider test abliteratedllm makes zero provider requests. All 15 seeds are listed, and there is no duplicate entry in the free directory.
Union guards on the merge with current dev: file-size ratchet, test layout, Spanish localization, preset counts, typecheck, and structure:check all passed.
Files also changed on dev since the branch base: tests/fixtures/test-layout-expected.json
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
ci:fullRun the four full Ubuntu test shards on this pull requestenhancementNew feature or request
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Disclosure: I run AbliteratedLLM.
AbliteratedLLM (https://abliteratedllm.dev) is a pay-per-token, OpenAI- and Anthropic-compatible API for abliterated (refusal-removed) and stock open-weight models, prepaid in USD by card or crypto.
This adds AbliteratedLLM as an ordinary API-key provider. It shows up in
ocx init, inocx provider add abliteratedllm, in the dashboard's Add provider picker and in API-key login. The preset reusesopenai-chatathttps://api.abliteratedllm.dev/v1. It adds no sponsor status, ranking preference, OAuth flow, adapter or wire pin. It follows the shape of cheapaimodel (#7899), Modelflare (#7037), AI Stupid Level (#7623) and TokenLab (#6221), and handles key validation the way OpenGateway does, because the model list is public.POST /v1/chat/completions, which serves every catalog model. That is the same wire the other OpenAI-compatible gateway presets use. The service also hasPOST /v1/responsesand an AnthropicPOST /v1/messages, but the preset doesn't use them. The Responses endpoint is stateless (storeis forced tofalseandprevious_response_idis rejected). A test pins Chat for all 15 models and all three inbounds.liveModels: false, owner decision 2026-11-18); the preset uses the seeded list below.GET /v1/modelsis public: on 2026-11-14 it returned 200 with no key and also with an invalid Bearer value. It can't verify a key, so the entry setsapiKeyValidation: "unknown", the same as OpenGateway.validateApiKeythen returnsunknownwithout making a request, so the login flow reports the key as unverifiable instead of giving a false positive. The service does have a key-gatedGET /v1/balance, which answers 401 for a missing or invalid key. The registry has no per-entry validation URL, though, and I didn't add one in this PR (see the end of this description).GET /v1/modelson 2026-11-14, in catalog order:glm-5.3-flash-abliterated(the default),deepseek-v4.1-flash-1017,deepseek-v4.1-flash-1017-ultrafast,deepseek-v4.1-flash-original,glm-5.3-flash,glm-5.3,glm-5.3-fast,kimi-k3,kimi-k3.1-pro,kimi-k3-fast,ember-1,muse-spark-open,qwen3.8-max,qwen3.8-2.4t-a95b,minimax-m3. This follows feat(providers): add cheapaimodel API-key preset #7899, which seeded its whole live catalog. The preset keeps this seeded list; live discovery is off.glm-5.3-flash-abliteratedis the service's abliterated model; the other 14 are the vendors' original weights.context_length: 1,048,576 tokens, except 262,144 for the two Qwen3.8 models and 512,000 forminimax-m3. Onlyglm-5.3-flash-abliteratedgets an output cap (131,072). Every other row listsmax_output_tokensequal to its window, so no cap is declared for them. Input modalities come frominput_modalities.glm-5.3andglm-5.3-fastare text only, so images for them go to the vision sidecar. The gateway would answer an image part sent to them with 400unsupported_content.glm-5.3-flashalso lists video, but it is seeded as text and image because the Codex catalog only carries text, image and audio. The seeds live inmodel-seeds.ts, as the Opper, AI Stupid Level and Morph catalogs do.reasoning_effortaslow,mediumorhighfor every model (some models also takenone). The preset therefore setsreasoningEfforts: ["low", "medium", "high"], which gives:xhighandmaxare sent ashighminimalis sent aslownoneomits the fieldpreserveCustomDestinationleaves an existing custom provider with the same name on its own URL.expected-prices.tsoverlays. Those overlays are for subscription and OAuth surfaces whose cost rows are missing, so none are added here. Prices are listed on the public model rows and at https://abliteratedllm.dev/pricing.GET /v1/modelsshowed on 2026-11-15. Of 15 models,glm-5.3-flash-abliteratedis the only abliterated one (variant: "abliterated"), and the other 14 haveprovider_label: "Stock model". None was labelled "Official DeepSeek API", the privacy policy's label for requests forwarded to DeepSeek's own API. The listed models, the DeepSeek ones included, run on Fireworks AI with zero data retention. Nothing is promised for models the service lists later. A test pins these sentences.This revision follows the owner's decision of 2026-11-15: a canonical preset, with the legal-entity evidence item waived. The free-directory reference row from the previous revision is dropped, so the service is listed once. The branch holds the two reviewed commits (
4bf2b3473,90c6e3f3a), rebased ondev04304da28on 2026-11-18. The guide text now lives on the API-key catalog page from #8768.Files:
src/providers/registry/entries-extended.tsabliteratedllmentry, aftercheapaimodel, with the routing notesrc/providers/registry/model-seeds.tstests/providers/abliteratedllm-provider.test.tsscripts/test-layout/layout.json,tests/fixtures/test-layout-expected.jsondocs-site/src/content/docs/guides/api-key-catalog.mddocs-site/src/content/docs/es/guides/api-key-catalog.mdguides/providers.md, all 9getting-started/quickstart.md,structure/ops/docs-and-release.mdPROVIDER_REGISTRYon currentdev: 109 total and 88 key-based (dev: 108 and 87)Evidence
Listed in the order
MAINTAINERS.mdgives. The endpoints were checked on 2026-11-14 and the routing on 2026-11-15.https://api.abliteratedllm.dev/v1/chat/completionswith a Bearer key. Streaming and function tools are supported.liveModels: false). It is public: it returned 200 without a key and with an invalid key on 2026-11-14. This is why the entry declaresapiKeyValidation: "unknown"and not the authenticated discovery that the contributing guide describes.invalid_api_keywithout a key and with an invalid key on 2026-11-14.Verification
Run with Bun 1.4.4 on Windows, after
bun install --frozen-lockfile(and ingui/, as CI does), first ondev1d10758fa. After the rebase onto04304da28(0 behind), I re-ran the first row (176 pass), the typecheck (no type errors in 2,566 files),structure:checkandprivacy:scan, and all passed.bun testoverabliteratedllm-provider,provider-registry-parity,free-directory-runcrate,docs-provider-preset-counts,docs-spanish-localization,file-size-ratchet,test-layoutandtest-layout-toolingbun testover all 65tests/providersandtests/ci-workflowsfiles that import the registry, plus the free-directory testsbun run typecheckbun run privacy:scan,structure:check,skill:surface:check,docs:clients:checkI didn't run the full suite (
bun run test), the changed-tests lane or the docs-site build on this revision; they are left to CI. On the earlier revision the docs-site build passed: 766 pages, 115,625 internal links. No live end-to-end test with a real AbliteratedLLM key was run. This change was prepared by an AI coding agent on my behalf.Maintainer update (2026-11-18)
Owner decision: the routing-authorization evidence item is waived, along with the legal-entity item waived on 2026-11-15. The preset uses its seeded model list (
liveModels: false), because the publicGET /v1/modelscannot validate a key. Maintainers pushed 6165691, which setsliveModels: false, updates the note and the en/es catalog paragraphs, and replaces the live-catalog test with a seeded-list test, and 0c3ba32, which mergesdev, keeps both the AbliteratedLLM and LLMTR layout registrations, and updates the seed comment.Checklist
https://api.abliteratedllm.dev/v1, and key validation never sends it anywhere because it reports unknown without a request).Review readiness checklist
This PR stays in draft until every box below is ticked. Tick all four boxes once the requirements are met:
Required local validation passed; commands, results, and any full-suite exception are documented.
I pushed my PR to a recent dev commit (at most 10 behind; a maintainer may still ask for the exact tip before merge).
I resolved all correct Codex and CodeRabbit findings.
My PR is ready for review.
Readiness boxes ticked by maintainer @lidge-jun at integration (2026-11-18) after the exact-head review and CI described in the PR comments.