This week's Insdr Link town hall focused on a practical expansion of the AI tools available to builders. We are adding a clearer OpenAI-compatible API surface for chat, vision, embeddings, and audio transcription, together with three model tiers for different kinds of engineering work.
All supported APIs are available through the public base URL:
https://api.insdrlink.com/v1
Builders should use only the supported public hostname and authenticate with their scoped Insdr Link API key.
New model tiers
We are adding three public model names so builders can choose the right balance for each task:
gpt-5.6-sol— our highest-capability tier for difficult reasoning, complex engineering, and vision work.gpt-5.6-terra— a balanced tier for everyday implementation, debugging, and agent workflows.gpt-5.6-luna— a lightweight tier for routine, high-volume tasks where speed and efficiency matter.
Using different tiers within the same workflow can make agentic development more efficient. Luna can handle repetitive subtasks, Terra can cover most implementation work, and Sol can be reserved for the problems that need deeper reasoning.
When creating an API key, set its models parameter to all proxy models rather than selecting only gpt-5.6-sol. The proxy now includes multiple chat tiers, embeddings, and transcription models, so a Sol-only key will not have access to the full set.
New model discovery endpoint
Builders can now verify which models their key can access with:
GET /v1/models
Example:
curl -fsS "https://api.insdrlink.com/v1/models" \
-H "Authorization: Bearer $LITELLM_API_KEY"
The response reflects the models allowed for that specific key.
Chat Completions endpoint
Text and multimodal conversations use:
POST /v1/chat/completions
The endpoint follows the familiar OpenAI Chat Completions format, making it straightforward to connect coding agents, scripts, and applications that support a custom OpenAI-compatible base URL.
curl -fsS "https://api.insdrlink.com/v1/chat/completions" \
-H "Authorization: Bearer $LITELLM_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-5.6-terra",
"messages": [
{"role": "user", "content": "Break this feature into verifiable implementation steps."}
]
}'
The GPT-5.6 model tiers also accept image URLs and base64 image data through Chat Completions. This enables tasks such as interface review, screenshot analysis, diagram interpretation, and visual debugging. Vision analyzes provided images; it does not generate new images.
Embeddings endpoint
Applications that need semantic search, retrieval, clustering, or RAG can use:
POST /v1/embeddings
The initial public embedding model is text-embedding-3-small, which returns 1,536-dimensional vectors.
curl -fsS "https://api.insdrlink.com/v1/embeddings" \
-H "Authorization: Bearer $LITELLM_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "text-embedding-3-small",
"input": ["First document", "Second document"]
}'
When building a vector index, store the model name and vector dimension alongside it. Changing either later requires existing content to be embedded again.
Audio transcription endpoint
Builders can now turn recorded speech into text with:
POST /v1/audio/transcriptions
The initial transcription model is gpt-4o-mini-transcribe.
curl -fsS "https://api.insdrlink.com/v1/audio/transcriptions" \
-H "Authorization: Bearer $LITELLM_API_KEY" \
-F "model=gpt-4o-mini-transcribe" \
-F "file=@recording.mp3"
Unlike the JSON-based chat and embeddings endpoints, audio transcription uses multipart/form-data. Builders should avoid uploading confidential recordings and should retain audio only as long as their application requires it.
OpenAI SDK compatibility
The endpoints can be used with the OpenAI SDK by setting the Insdr Link base URL and a scoped API key:
import os
from openai import OpenAI
client = OpenAI(
api_key=os.environ["LITELLM_API_KEY"],
base_url="https://api.insdrlink.com/v1",
)
response = client.chat.completions.create(
model="gpt-5.6-luna",
messages=[{"role": "user", "content": "Reply with OK"}],
)
print(response.choices[0].message.content)
This compatibility makes it possible to reuse many existing tools without adopting a proprietary client library.
API key setup reminder
When creating an API key, set the models parameter to all proxy models. Do not configure the key for only gpt-5.6-sol; the proxy now includes several models, and selecting all proxy models ensures the key can use the newly added options.
What this unlocks
These additions give Insdr Link builders one consistent API surface for several common project needs:
- coding and reasoning workflows;
- vision-assisted analysis;
- semantic search and RAG;
- voice prompting and transcription;
- integrations with OpenAI-compatible tools and agents.
We are sponsoring AI and compute so engineers can spend more time building and testing real projects. This week's changes expand the kinds of systems builders can create while keeping setup simple: one public base URL, scoped access, and model tiers suited to different workloads.