Gradient AI API
67
AI ReadinessAI-Aware (B)67/100
Official vendor OpenAPI document · agent-readyAI/MLLanguage Modelsbearer22 EndpointsREST
For Agents
Run LLM completions, fine-tune models, generate embeddings, manage RAG collections, and call prebuilt AI blocks for sentiment, entities, summarisation, Q&A, and PDF extraction.
Use for: Generate a chat completion with a Gradient model, Fine-tune a model on a customer support dataset, Create vector embeddings for a list of documents, Set up a RAG collection over a folder of PDFs
Not supported: Does not handle image generation, audio synthesis, or hosting third-party models - use for LLM completion, fine-tuning, embeddings, RAG, and prebuilt task blocks only.
Gradient AI is an LLM platform that combines model fine-tuning, completion and embedding generation, retrieval-augmented generation collections, and a library of prebuilt AI building blocks. Its 22-endpoint API covers the full lifecycle: list and create models, fine-tune them, run completions, generate embeddings, manage RAG collections and their indexed files, and call task-specific blocks for sentiment, entity extraction, summarisation, question answering, PDF extraction, and personalisation. Authentication is HTTP Bearer.
Install Jentic One Beta
Connect the Gradient AI API to your agent
Jentic One is a self-hosted execution layer for AI agents. It lets your agent call the Gradient AI API, or any other public or private API you need. You set the rules, the agent never sees your credentials, and every call is logged.
Two steps, two machines. Install the instance in a safe environment, then register your agent from wherever it runs.
1
Step 1: Jentic One Host machine
# On the machine that will host your Jentic One instance:
curl -fsSL "https://jentic.com/install.sh?src=apis&api=%2Fapis%2Fgradient.ai%2Fgradient-ai" | sh2
Step 2: Agent machine
# On the machine where your agent runs (keep this separate from the instance):
curl -fsSL "https://jentic.com/install.sh?src=apis&api=%2Fapis%2Fgradient.ai%2Fgradient-ai" | sh
jentic register # connects your agent to your Jentic One instanceJentic One is in public beta. The setup above keeps your agent separate from the instance, which is what you want before using real credentials: an agent running as the same OS user as Jentic One can read its stored keys directly. Just evaluating? A single local install is fine to start. See the secure deployment guide for the tiers.
Capabilities
What an agent can do with Gradient AI API.
Generate completions from a model with POST /models/{id}/complete
Fine-tune a base model on uploaded data through POST /models/{id}/finetune
Create and query embeddings via POST /embeddings
Build and populate RAG collections with /rag-collections and /rag-collections/{id}/files
Call prebuilt blocks for sentiment, entity extraction, summarisation, Q&A, and PDF extraction
Personalise documents for a target audience with POST /blocks/personalize
Upload files for downstream fine-tuning, RAG, or block input via POST /files
Use Cases
Patterns agents use Gradient AI API for, with concrete tasks.
★ Domain-Specific Fine-Tuning
Teams with proprietary data fine-tune a Gradient base model on customer support transcripts or domain corpora. POST /files uploads the training data, POST /models creates the new model, and POST /models/{id}/finetune kicks off the job. The resulting model is then served through the same /models/{id}/complete endpoint as base models, keeping inference code unchanged.
POST /files with a JSONL training set, POST /models to register the target model, then POST /models/{id}/finetune and poll until status reaches completion.
Retrieval-Augmented Generation Over Private Docs
POST /rag-collections creates a collection, POST /rag-collections/{id}/files indexes uploaded PDFs and HTML, and downstream completion calls scoped to the collection ground responses in the indexed material. This avoids hallucination on proprietary content and removes the need for a separate vector store. The agent still chooses which model handles the final completion.
Create a RAG collection, upload 50 PDFs via POST /rag-collections/{id}/files, then call POST /blocks/answer scoped to that collection for a user question.
Building Block Pipelines
Rather than write prompts from scratch, agents call prebuilt blocks: /blocks/sentiment for tone, /blocks/entity for entities, /blocks/summarize for summaries, /blocks/answer for grounded Q&A, /blocks/pdf for PDF extraction, and /blocks/personalize for tailored content. Each block returns a structured payload, making them easy to chain inside an agent's reasoning loop.
POST /blocks/pdf with a contract upload, then POST /blocks/entity on the extracted text to surface party names and dates.
Embedding Generation for Search
POST /embeddings turns a list of texts into vector embeddings used to power semantic search inside the agent's own pipeline. GET /embeddings lists available embedding models so the agent can pick a dimensionality and language match. This is for cases where the team owns the vector store and only wants embeddings as a service rather than a full RAG pipeline.
POST /embeddings with 200 article titles and store the returned vectors in pgvector for cosine search.
Agent Integration via Jentic
An agent that wants 'a fine-tuned customer support model with RAG over our help centre' uses Jentic to discover Gradient AI, walk through /files upload, /models creation, /models/{id}/finetune, /rag-collections, and finally /models/{id}/complete. The Bearer token sits in Jentic's vault, so credentials never enter the agent's prompt context.
Use the Jentic SDK to search 'fine tune a model and run rag completion', chain /files, /models, /models/{id}/finetune, /rag-collections, and /models/{id}/complete in sequence.
Key Endpoints
22 endpoints — gradient ai is an llm platform that combines model fine-tuning, completion and embedding generation, retrieval-augmented generation collections, and a library of prebuilt ai building blocks.
METHOD
PATH
DESCRIPTION
POST
/models/{id}/complete
Generate a completion from a model
POST
/models/{id}/finetune
Fine-tune a model on uploaded data
POST
/embeddings
Create embeddings from a list of texts
POST
/rag-collections
Create a RAG collection
POST
/rag-collections/{id}/files
Add files to a RAG collection
POST
/blocks/summarize
Summarise a document
POST
/blocks/answer
Answer a question grounded in a document
POST
/blocks/pdf
Extract structured content from a PDF
POST
/models/{id}/complete
Generate a completion from a model
POST
/models/{id}/finetune
Fine-tune a model on uploaded data
POST
/embeddings
Create embeddings from a list of texts
POST
/rag-collections
Create a RAG collection
POST
/rag-collections/{id}/files
Add files to a RAG collection
POST
/blocks/summarize
Summarise a document
POST
/blocks/answer
Answer a question grounded in a document
POST
/blocks/pdf
Extract structured content from a PDF
Jentic AI Readiness Score
This API is usable in Jentic One now. Its AI-readiness score against Jentic's framework shows where it stands today and where improvements would make it even easier for agents to use.
G
Gradient AI API
- AI-Aware (B)67/100
93
Foundational Compliance63
Developer Experience & Jentic Compatibility50
AI-Readiness & Agent Experience94
Agent Usability60
Security69
AI DiscoverabilityPowered by JenticScoring Framework 1.0.0 | Scoring Engine 0.4.0
Show dimension breakdown
93
Foundational Compliance
Base layer of spec validity and structural soundness.
Grade: A+Signals: 4
98%
Lint Results
Aggregated quality score from linter diagnostics, weighted by severity.
100%
Resolution Completeness
Percentage of `$ref` references that resolve successfully.
100%
Specification Validity
Checks whether the API description parses successfully and conforms to its declared specification (e.g., OpenAPI).
75%
Structural Integrity
Structural correctness score based on schema issues using logarithmic dampening.
63
Developer Experience & Jentic Compatibility
Clarity, completeness, and ingestion readiness for developers and tooling.
Grade: B-Signals: 4
0%
Example Density
How richly the API is illustrated with examples.
100%
Example Validity
Percentage of examples that conform to their schemas.
50%
Response Coverage
Percentage of operations with complete response definitions (success, client error, server error).
100%
Tooling Readiness
Health of API ingestion, bundling, and resolution within Jentic pipelines.
50
AI-Readiness & Agent Experience
Semantic breadth, depth, and agent comprehension for AI systems.
Grade: C-Signals: 4
46%
Description Coverage
Coverage of descriptions across API elements.
0%
Error Standardization
Coverage of RFC 9457 Problem Details for error responses.
100%
OperationId Quality
Coverage, uniqueness, and casing consistency of operationIds for AI inference.
56%
Summary Coverage
Coverage of summaries across operations/tags/info.
94
Agent Usability
Functional utility, complexity comfort, and AI orchestration readiness.
Grade: A+Signals: 1
94%
Complexity Comfort
Agent comfort level based on API operational and structural complexity.
60
Security
Trust, risk posture, and security compliance.
Grade: B-Signals: 1
60%
Authentication Strength
Average quality of security schemes based on authentication method strength (weakest link for OAuth2).
69
AI Discoverability
Findability, semantic richness, and reasoning readiness.
Grade: B+Signals: 1
69%
Descriptive Richness
Clarity and depth of descriptions across API elements.
Score it yourself
Every API in the directory is allowlisted, so you can re-score it with no key required.
npx @jentic/api-scorecard-cli score <openapi-url>Why Jentic?
What agents get from Jentic-routed access to this vendor.
Setup
Wiring Gradient AI by hand means setting up its bearer auth, tracking which model id goes in each completion or fine-tune call, and juggling separate completion, embedding, RAG, and prebuilt block endpoints yourself. Through Jentic you install once, import the Gradient AI API from the API Directory, store the bearer token once, and your agent calls it.
Permission scoping
Gradient AI puts the model id in the URL path (/models/{id}/complete, /models/{id}/finetune), so a rule can pin your agent to one model: it can run completions and embeddings for that model and nothing else. You choose the operations it may call, so destructive ones like starting a fine-tune or deleting a RAG collection are not included unless you add them.
Credential isolation
Your Gradient AI bearer token is stored once, encrypted, by your own Jentic One instance and injected at execution time. It never enters the agent's prompt, logs, or context.
Intent-based discovery
Agents search Jentic by intent such as 'get an embedding' or 'summarize a document with a prebuilt block', and Jentic returns the matching Gradient AI operation with its input schema so the agent calls the right endpoint without browsing the reference docs.
Related APIs
Alternatives and complements available in the Jentic catalogue.
FAQs
Specific to using Gradient AI API through Jentic.
What authentication does the Gradient AI API use?
Gradient AI uses HTTP Bearer authentication (security scheme 'bearerAuth'). Issue a key in the Gradient AI dashboard and send it as Authorization: Bearer <token>. Through Jentic, the bearer token is stored in your Jentic One instance and injected at execution time so the agent never sees the raw secret.
Can I fine-tune my own model on Gradient AI?
Yes. POST /models creates a new model entry, POST /files uploads training data, and POST /models/{id}/finetune starts the job. After completion the resulting model is callable through POST /models/{id}/complete using the same shape as base models.
What are the rate limits for the Gradient AI API?
The OpenAPI spec does not declare per-endpoint rate limits. Limits are tied to your plan tier and to specific endpoints (completion, fine-tune, embeddings) which have separate concurrency caps. Use the Gradient AI dashboard to monitor usage.
How do I run a RAG completion through Jentic?
Search Jentic for 'create a rag collection and answer a question', chain POST /rag-collections, POST /rag-collections/{id}/files for indexing, and POST /blocks/answer to ask a grounded question. The Jentic SDK returns each step's response so the agent can pass ids forward.
What are 'blocks' in the Gradient AI API?
Blocks are prebuilt task endpoints - sentiment, entity, summarize, answer, pdf, and personalize. Each accepts a domain-specific input (text, document, persona) and returns structured output, so the agent does not need to author prompts from scratch for these tasks.
Is the Gradient AI API free?
Gradient AI offers a developer tier with limited monthly quota and paid tiers with higher concurrency, fine-tuning, and dedicated capacity. Pricing is published on gradient.ai and tiers are enforced at the API gateway.
Can I limit what my agent is allowed to do with the Gradient AI API?
Yes. Jentic One runs on infrastructure you host, so your own rules decide which Gradient AI operations and credentials the agent may use. Because Gradient AI puts the model id in the URL path (/models/{id}/complete, /models/{id}/finetune), you can pin the agent to a single model and to read-style calls like running completions and generating embeddings, while leaving out destructive operations such as starting a fine-tune or deleting a RAG collection unless you explicitly add them. The bearer token stays stored, encrypted, in your own instance and is injected only at execution time, so the agent can only reach the endpoints you allowed.
GET STARTED