Document Tuning Engines OpenAI-compatible configuration#5981
Conversation
📝 WalkthroughWalkthroughThis PR adds a new documentation example demonstrating how to configure an ChangesTuning Engines OpenAI-compatible endpoint documentation
Estimated code review effort🎯 1 (Trivial) | ⏱️ ~3 minutes
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
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. Comment |
There was a problem hiding this comment.
Actionable comments posted: 1
🧹 Nitpick comments (1)
docs/en/learn/llm-connections.mdx (1)
160-161: 💤 Low valueConsider a more neutral comment tone.
The comment lists specific features ("model routing, policy controls, audit logs...") which reads somewhat promotional. Compare with the Google example on line 171 which uses a simpler, more neutral description:
# Example using Gemini's OpenAI-compatible API.Consider simplifying to something like:
# Example using Tuning Engines OpenAI-compatible endpointThis aligns better with the documentation style of other examples while still being descriptive.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@docs/en/learn/llm-connections.mdx` around lines 160 - 161, Replace the promotional comment line beginning with "# Governed OpenAI-compatible endpoint for model routing, policy controls, audit logs, traces, approvals, and cost visibility." with a shorter, more neutral comment such as "# Example using Tuning Engines OpenAI-compatible endpoint" so it matches the tone of the other examples (e.g., the Google/Gemini example) and keeps the description concise and factual.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@docs/en/learn/llm-connections.mdx`:
- Around line 159-168: Update the Tuning Engines example in the LLM snippet to
use an actual supported model ID and neutral wording: replace
model="gpt-4o-mini" with a verified Tuning Engines model ID (or cite the exact
model/source in a short parenthetical), keep the LLM(...) and Agent(...) usage
but ensure base_url="https://api.tuningengines.com/v1" remains consistent with
an OpenAI-compatible gateway, and change the leading comment to a neutral
instructional sentence (e.g., "Example showing how to configure an
OpenAI-compatible gateway for model routing and policy controls") instead of
promotional language; ensure references to the LLM class, model parameter,
base_url, and Agent class are updated accordingly.
---
Nitpick comments:
In `@docs/en/learn/llm-connections.mdx`:
- Around line 160-161: Replace the promotional comment line beginning with "#
Governed OpenAI-compatible endpoint for model routing, policy controls, audit
logs, traces, approvals, and cost visibility." with a shorter, more neutral
comment such as "# Example using Tuning Engines OpenAI-compatible endpoint" so
it matches the tone of the other examples (e.g., the Google/Gemini example) and
keeps the description concise and factual.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro Plus
Run ID: 12745f3c-60bf-44fc-acc2-b47422c1ca6b
📒 Files selected for processing (1)
docs/en/learn/llm-connections.mdx
| ```python Tuning Engines | ||
| # Governed OpenAI-compatible endpoint for model routing, | ||
| # policy controls, audit logs, traces, approvals, and cost visibility. | ||
| llm = LLM( | ||
| model="gpt-4o-mini", | ||
| api_key="your-tuning-engines-key", | ||
| base_url="https://api.tuningengines.com/v1" | ||
| ) | ||
| agent = Agent(llm=llm, ...) | ||
| ``` |
There was a problem hiding this comment.
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
# Description: Verify the Tuning Engines endpoint is accessible
# Check if the endpoint is publicly accessible
curl -I -s -o /dev/null -w "%{http_code}" https://api.tuningengines.com/v1 || echo "Failed to connect"Repository: crewAIInc/crewAI
Length of output: 63
🌐 Web query:
Tuning Engines API base URL documentation
💡 Result:
Tuning Engines API base URL (from available integration docs/examples): https://api.tuningengines.com/v1 [1][2] Evidence notes: - Multiple community/developer integration examples reference the base URL as https://api.tuningengines.com/v1 (e.g., OpenAI-compatible setup examples mention “Base URL: https://api.tuningengines.com/v1”). [1][3] - The tuning-engines CLI documents a configurable “TE_API_URL” with a different default of https://app.tuningengines.com (this appears to be the CLI/UI backend, not the public versioned API gateway). [4][5] Practical guidance: - If you are calling the versioned REST/OpenAI-compatible API directly, use https://api.tuningengines.com/v1 as the base URL. [1][3] - If you are using the tuning-engines CLI, set TE_API_URL explicitly if your environment expects the API gateway; otherwise it defaults to https://app.tuningengines.com. [4][5]
Citations:
- 1: docs: add Tuning Engines provider example anomalyco/opencode#28533
- 2: https://tuningengines.com/
- 3: docs: add Tuning Engines OpenAI-compatible setup cline/cline#10938
- 4: https://github.com/cerebrixos-org/tuning-engines-cli
- 5: https://registry.npmjs.org/tuningengines-cli
🌐 Web query:
Tuning Engines supported OpenAI models
💡 Result:
As of 2026-05-30, the OpenAI fine-tuning API supports creating fine-tuning jobs only for specific base models (and the set is shrinking as OpenAI winds down fine-tuning). The base models explicitly referenced by OpenAI for fine-tuning include: 1) gpt-4o-mini-2024-07-18 [1] 2) gpt-4o-2024-08-06 [2] 3) gpt-4.1-2025-04-14 [3] 4) gpt-4.1-mini-2025-04-14 [3] 5) gpt-4.1-nano-2025-04-14 [3] 6) o4-mini (reinforcement fine-tuning method is supported, and OpenAI notes o4-mini for RL fine-tuning availability) [4] 7) gpt-3.5-turbo (legacy name appears as selectable in the fine_tuning resource reference) [5] 8) babbage-002 (legacy name appears as selectable in the fine_tuning resource reference) [5] 9) davinci-002 (legacy name appears as selectable in the fine_tuning resource reference) [5] Important constraints (time sensitivity): - OpenAI is “winding down” the fine-tuning platform; it is no longer accessible to new users, while existing users can create training jobs for coming months [3]. - OpenAI provides a deprecations page with shutdown dates and notes on training-job availability restrictions (e.g., creation restrictions for orgs that have not previously fine-tuned, and later hard shutdown for new fine-tune jobs) [6]. Note: “Tuning Engines” are distinct from OpenAI fine-tuning. The term “Tuning Engines supported OpenAI models” can also mean third-party services (like Tuning Engines) that tune open models and may support “OpenAI-compatible” endpoints; that’s not the same as OpenAI’s fine-tuning API base-model support. The official OpenAI fine-tuning support is the list above from OpenAI’s docs and references [5][3][1][2].
Citations:
- 1: https://developers.openai.com/cookbook/examples/how_to_finetune_chat_models
- 2: https://openai.com/index/gpt-4o-fine-tuning/
- 3: https://developers.openai.com/api/docs/guides/fine-tuning-best-practices
- 4: https://community.openai.com/t/openai-is-winding-down-the-fine-tuning-api-and-platform-discussion-thread/1380522
- 5: https://developers.openai.com/api/reference/resources/fine_tuning/
- 6: https://developers.openai.com/api/docs/deprecations
🏁 Script executed:
#!/bin/bash
# Check if the base URL behaves like a valid OpenAI-compatible gateway (should be 401/403/404 depends on auth)
set -euo pipefail
BASE="https://api.tuningengines.com/v1"
for path in "" "/models" "/chat/completions" "/responses"; do
url="${BASE}${path}"
echo "== $url =="
curl -sS -o /dev/null -w "HTTP_CODE=%{http_code}\n" -I "$url" || echo "curl_failed"
doneRepository: crewAIInc/crewAI
Length of output: 306
🌐 Web query:
Tuning Engines gpt-4o-mini supported models OpenAI compatible base_url /v1
💡 Result:
Tuning Engines gpt-4o-mini supported models, OpenAI-compatible base_url (/v1) I couldn’t find any authoritative/official Tuning Engines documentation stating (a) which exact base/fine-tunable model IDs (including gpt-4o-mini variants) are supported and (b) whether their platform exposes an OpenAI-compatible API with a base_url that includes /v1. What I did find (related, but not the requested “Tuning Engines” API compatibility): 1) Tuning Engines is described as a managed fine-tuning platform for (open-source) models via their agents/CLI, not as an OpenAI-compatible endpoint for OpenAI-hosted models. Their site and CLI emphasize fine-tuning open models like Qwen, Llama, DeepSeek, Mistral, Gemma, etc. [1][2] 2) OpenAI’s own API docs show GPT-4o-mini fine-tuning availability for the base model name gpt-4o-mini and the snapshot gpt-4o-mini-2024-07-18. [3][4] 3) For OpenAI-compatible clients, the “/v1” base path is standard for OpenAI’s API (e.g., POST https://api.openai.com/v1/fine_tuning/jobs). [4][5] 4) For OpenAI-compatible clients targeting Azure OpenAI, Microsoft’s tutorial shows base_url formatted as https://.openai.azure.com/openai/v1/ when using the OpenAI SDK. [6][7] Practical takeaway: - If you mean “OpenAI API compatibility for fine-tuning GPT-4o-mini (gpt-4o-mini-2024-07-18)”, that is documented by OpenAI and Azure OpenAI, but not by Tuning Engines in the sources I found. [3][4][6] - If you mean “Tuning Engines’ own supported base models” and “whether it supports OpenAI-compatible base_url /v1”, I need you to point me to Tuning Engines’ API/MCP docs (or tell me the exact endpoint/product name). Right now, the available sources don’t include that level of detail. [1][2] If you share a link to Tuning Engines’ API reference (or tell me what endpoint they provide), I can re-run the research specifically to extract: - supported base model IDs (including whether gpt-4o-mini-2024-07-18 is supported) - the exact OpenAI-compatible base_url (e.g., ending in /v1) and the fine-tuning route naming (/fine_tuning/jobs vs something else).
Citations:
- 1: https://tuningengines.com/
- 2: https://github.com/cerebrixos-org/tuning-engines-cli
- 3: https://developers.openai.com/api/docs/models/gpt-4o-mini
- 4: https://developers.openai.com/api/reference/resources/fine_tuning/
- 5: https://openai-hd4n6.mintlify.app/api-reference/fine-tuning/create-a-fine-tuning-job
- 6: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/tutorials/fine-tune
- 7: https://learn.microsoft.com/en-us/azure/foundry-classic/openai/tutorials/fine-tune
🌐 Web query:
site:tuningengines.com "gpt-4o-mini" "model"
💡 Result:
I’m not able to produce a source-grounded answer for “gpt-4o-mini” (e.g., its official model documentation, release details, or API model name) from the evidence available in the searches I ran: the only result returned was unrelated (a third-party site about fine-tuning open models) and the subsequent queries for authoritative OpenAI/platform documentation returned no results. If you want, share a link you’re expecting me to use (OpenAI model card / API docs / release notes), or allow me another research pass with more targeted queries (e.g., including “site:platform.openai.com gpt-4o-mini”, “site:help.openai.com gpt-4o-mini”, or “models list gpt-4o-mini”).
Citations:
🏁 Script executed:
#!/bin/bash
set -euo pipefail
BASE="https://api.tuningengines.com/v1"
echo "## GET /v1/models (no auth)"
curl -sS -o /dev/null -w "HTTP_CODE=%{http_code}\n" -X GET "$BASE/models" || echo "curl_failed"
echo "## POST /v1/chat/completions (no auth, minimal body)"
curl -sS -o /dev/null -w "HTTP_CODE=%{http_code}\n" \
-X POST "$BASE/chat/completions" \
-H "Content-Type: application/json" \
--data '{"model":"gpt-4o-mini","messages":[{"role":"user","content":"hi"}]}' || echo "curl_failed"
echo "## POST /v1/responses (no auth, minimal body)"
curl -sS -o /dev/null -w "HTTP_CODE=%{http_code}\n" \
-X POST "$BASE/responses" \
-H "Content-Type: application/json" \
--data '{"model":"gpt-4o-mini","input":"hi"}' || echo "curl_failed"Repository: crewAIInc/crewAI
Length of output: 229
🌐 Web query:
site:tuningengines.com openai-compatible chat/completions responses base_url v1
💡 Result:
The standard endpoint for OpenAI-compatible chat completions is /v1/chat/completions [1]. When building or integrating with systems that offer OpenAI-compatible APIs, the base_url generally refers to the root path of the service provider's API server, often ending before the /v1 versioning prefix [2][1]. To interact with these services using standard OpenAI client SDKs (such as the official OpenAI Python library), you typically configure the client by providing the specific host URL as the base_url (or base_url parameter in newer SDK versions) [1]. For example, if an API provider hosts their service at api.example.com, the full request path for the completions endpoint is constructed as: https://api.example.com/v1/chat/completions When initializing an OpenAI-compatible client, you would set the base_url to: https://api.example.com/v1 This allows the client to correctly append the /chat/completions path for requests [1]. Always verify the specific base URL requirements in the documentation of the platform you are integrating with, as some may expect the base_url to terminate at the root domain while others may include the /v1 prefix [1].
Citations:
🌐 Web query:
site:github.com tuning-engines openai-compatible base_url /v1 responses
💡 Result:
Tuning Engines is a platform that provides specialized fine-tuning services for various open-source LLMs (such as Llama, Qwen, DeepSeek, Mistral, and others) [1]. To integrate Tuning Engines into applications or AI assistants that expect an OpenAI-compatible API, users configure the base URL to point to the Tuning Engines API endpoint and use a specific API key format [2]. For OpenAI-compatible integrations, the standard base URL is: https://api.tuningengines.com/v1 [2] Users typically authenticate with an API key formatted as sk-te-... [2]. Many AI development tools and agent frameworks (such as Continue or Cline) support this by allowing users to set a custom base URL and API key [3][4][2][5]. When configured this way, the application routes its requests—such as chat completions or model interactions—to the Tuning Engines infrastructure instead of the default OpenAI servers [6]. Tuning Engines provides a CLI and MCP (Model Context Protocol) server to manage fine-tuning jobs, track model status, and handle GPU orchestration [1]. The platform defaults to https://app.tuningengines.com for general API operations [1], while the /v1 path is specifically used for OpenAI-compatible model inference and completion endpoints [2].
Citations:
- 1: https://github.com/cerebrixos-org/tuning-engines-cli
- 2: docs: add Tuning Engines OpenAI-compatible setup cline/cline#10938
- 3: docs: add Tuning Engines OpenAI-compatible example continuedev/continue#12447
- 4: docs: add Tuning Engines provider example anomalyco/opencode#28533
- 5: https://github.com/continuedev/continue/actions/runs/26193224834
- 6: vercel/ai@9a51b92
🌐 Web query:
site:tuningengines.com "gpt-4o-mini"
💡 Result:
As of May 30, 2026, there is no widely recognized or official information available regarding a model specifically designated as "gpt-4o-mini" from OpenAI. Current OpenAI model offerings include the GPT-4o series, but the specific "mini" designation does not appear in current authoritative documentation or broad public records [1][2]. The term may be a misunderstanding or a reference to a model that has not been officially released or documented by OpenAI as of this date.
Citations:
🌐 Web query:
site:github.com tuning-engines-cli "gpt-4o-mini"
💡 Result:
The query “tuning-engines-cli "gpt-4o-mini"” appears to refer to the GitHub project cerebrixos-org/tuning-engines-cli, which is described as a “CLI & MCP server for Tuning Engines” that can fine-tune open LLMs on code repositories via a CLI or an MCP-compatible AI assistant. [1] However, the available sources I found do not explicitly show that this specific CLI accepts or supports the OpenAI model ID “gpt-4o-mini” as a tuning target. In other words: I can confirm what the tuning-engines-cli project is and that it supports fine-tuning open models (e.g., Llama/Qwen/etc.), but I cannot confirm from cited sources that you can fine-tune “gpt-4o-mini” specifically with it. [1] What I can support with evidence from the tuning-engines-cli repository: 1) Purpose: It’s a CLI/MCP server for “Tuning Engines” to tailor/fine-tune open models on user data/repositories. [1] 2) Training options/features at a high level: it mentions support for LoRA/QLoRA/full fine-tuning and managed GPU provisioning/training orchestration. [1] 3) Job-related MCP/CLI tool surface: it lists operations like create_job, estimate_job, list_jobs, show_job, cancel_job, retry_job, and model import/export/status functions. [1] There is another unrelated project in results that mentions “gpt-4o-mini” (wkdkavishka/GPT-shell-4o-mini), but that is a terminal client for using OpenAI models (not “tuning-engines-cli” nor tuning/fine-tuning). [2] If you want, paste the exact command you’re trying to run (or the CLI help output), and I can determine whether “gpt-4o-mini” is meant to be a supported model parameter for tuning-engines-cli, or whether you should instead use one of the open-model families it supports (as suggested by the repository description).
Citations:
- 1: https://github.com/cerebrixos-org/tuning-engines-cli
- 2: https://github.com/wkdkavishka/gpt-shell-4o-mini
🌐 Web query:
site:tuningengines.com supported models llama qwen deepseek mistral model id
💡 Result:
On tuningengines.com, the publicly listed “Model library” includes these supported open-model families/variants: Llama (3.3 70B, 3.1 8B, 3.2 Vision), DeepSeek (V3, R1), Qwen (2.5 72B, 2.5 Coder 32B), and Mistral (Mistral Small 3, Mixtral 8x7B). [1]
The site does not publish the exact API model ID strings for those entries (it only shows an example call using model="gpt-4o-mini" and notes an OpenAI-compatible base URL). To get the exact IDs, you’d typically need to query the provider’s models list from their API/dashboard. [1]
Update the Tuning Engines example to ensure model is actually supported
- In
docs/en/learn/llm-connections.mdx(lines 159-168),base_url="https://api.tuningengines.com/v1"is consistent with an OpenAI-compatible gateway: unauthenticated requests to/models,/chat/completions, and/responsesreturn401(not404). model="gpt-4o-mini"still isn’t supported by a clearly published “supported model IDs” list in the sources found; align the snippet with Tuning Engines’ real model IDs (or cite their exact model/source) to avoid readers copying a non-workingmodelvalue.- The comment wording is more promotional than the neighboring examples; make the description more neutral/instructional.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@docs/en/learn/llm-connections.mdx` around lines 159 - 168, Update the Tuning
Engines example in the LLM snippet to use an actual supported model ID and
neutral wording: replace model="gpt-4o-mini" with a verified Tuning Engines
model ID (or cite the exact model/source in a short parenthetical), keep the
LLM(...) and Agent(...) usage but ensure
base_url="https://api.tuningengines.com/v1" remains consistent with an
OpenAI-compatible gateway, and change the leading comment to a neutral
instructional sentence (e.g., "Example showing how to configure an
OpenAI-compatible gateway for model routing and policy controls") instead of
promotional language; ensure references to the LLM class, model parameter,
base_url, and Agent class are updated accordingly.
Summary
Why this belongs here
Tuning Engines is an AI control plane for teams that want centralized model access, routing, tenant-scoped keys, policy/guardrail checks, audit logs, traces, approvals, and usage/cost visibility around existing AI applications.
This project already supports OpenAI-compatible endpoints. The docs change makes that existing capability discoverable for users who want governance and observability without rewriting their application around a separate SDK or changing this project’s runtime behavior.
Testing
git diff --check