evren-platform-api is a free, open-source, unofficial guide and toolkit for the EVREN AI platform's REST API that shows how to create datasets, upload files resumably and publish them, with a ready OpenCode workflow.
English | Türkçe
Unofficial. This project is not affiliated with or endorsed by SSB or the EVREN platform. "EVREN" is used only to say which API the code works with. No EVREN logo, colours or branding are used. You act under your own account and under the platform's terms; uploaded datasets and models may be used by the platform for training under its user agreement, so read that agreement before you upload.
Why this exists
Community projects already cover EVREN's LLM gateway: an OpenCode provider installer, command-line clients, chat UIs, and an MCP server for training. What nobody had written down is the Platform REST API for datasets:
- which API key type can call which endpoint,
- that an NLP dataset can be created through the API even though the web form only offers vision and medical,
- the three-step upload (presign, PUT, confirm), how deduplication works, and how to verify the result,
- how to do all of it resumably, from a script or from an AI coding agent.
This repository documents those facts as they were measured against the live API on 2026-10-26, and ships one small tool that does the work.
Quick start
You need Python 3.9 or newer. There is nothing to install.
1. Get a key. In the EVREN web app open Settings -> API Keys and create an evren_plt_ key (it is shown once).
git clone https://github.com/nsozturk/evren-platform-api
cd evren-platform-api
export EVREN_PLATFORM_KEY="evren_plt_..." # your own key; never commit it
2. Prepare a dataset folder with your data files, a dataset card and a README:
mkdir my-dataset # put train.parquet, data.jsonl, ... in it
cp examples/card.example.json my-dataset/card.json # edit name, description, license, tags
$EDITOR my-dataset/README.md # the dataset card shown on the platform
3. Create the dataset (private NLP; preview first):
python3 tools/evren_dataset_upload.py manifest my-dataset
python3 tools/evren_dataset_upload.py create --card my-dataset/card.json --readme my-dataset/README.md --dry-run
python3 tools/evren_dataset_upload.py create --card my-dataset/card.json --readme my-dataset/README.md --visibility private
4. Upload. Safe to stop and re-run at any point; it resumes.
python3 tools/evren_dataset_upload.py upload my-dataset
5. Verify that every file is on the server with a matching SHA-256:
python3 tools/evren_dataset_upload.py verify my-dataset
# verify my-dataset: server 9 / manifest 9 | missing [] | extra [] | not ready [] -> OK
The uploader prints timestamped progress lines, retries on 5xx, 429 and network errors with backoff, streams files from disk instead of loading them into memory, and keeps a journal so a re-run skips finished files. Defaults live inside the dataset folder (.evren_dataset.json, .evren_upload_journal.jsonl, .evren_upload.log). Exit codes: 0 ok, 1 operation failed, 2 usage or configuration error.
Which API key do I need?
| Key | Prefix | Host | Header | Works for |
|---|
| Full platform key | evren_plt_ | https://api.ssyz.org.tr/api/v1 | X-API-Key | Datasets, models, billing, catalogue |
| Models-only key | evren_mdl_ | same | X-API-Key | Models and inference only (dataset calls return 403) |
| LLM gateway key | evren_llm_ | https://evren-llmapi.ssyz.org.tr/v1 | Authorization: Bearer | Chat, embeddings, OCR, ... (not the Platform API) |
Dataset work needs evren_plt_. Details and the measured scope matrix: docs/api/keys.md.
API reference
| Page | Covers |
|---|
| docs/api/keys.md | Hosts, headers, key types and scopes, response envelope |
| docs/api/datasets.md | Create, list, delete, items, catalogue search, versions, export |
| docs/api/upload.md | Presign / PUT / confirm with curl and Python, dedup, verification, sharding, .npy to Parquet |
| docs/api/llm-gateway.md | Short orientation to the LLM gateway and its terms flow |
| docs/api/errors.md | Error shape, the 401/403 cases, retry guidance, exit codes |
| docs/opencode.md | OpenCode provider block and the repo's OpenCode command |
Use it with OpenCode
OpenCode is the first supported agent harness. The repository contains an AGENTS.md contract and a custom command, so an agent can publish a folder for you:
export EVREN_PLATFORM_KEY="evren_plt_..."
opencode
# inside OpenCode:
/evren-publish-dataset ./my-dataset
The command builds the manifest, creates a private NLP dataset, uploads, verifies, reports, and asks before anything public. To use EVREN-hosted models as your OpenCode model, add the provider block from opencode.example.json (key from EVREN_LLM_API_KEY); see docs/opencode.md. To give any assistant these instructions globally, paste GLOBAL_AGENT_RULE.md into its global instructions.
Platform credits
From the platform's pricing page (2026-10-26; check it for current numbers):
| Item | Credits (CR) |
|---|
| Starting balance | 1000 |
| Daily login | +10 |
| Publishing a dataset | +5 (up to 3 per day) |
| Item milestones at 100 / 500 / 1K / 5K / 10K | +2 / +5 / +15 / +25 / +40 |
| Daily earning cap | 100 |
| GPU training, 1x H200 | 10 per hour |
Milestones count files, not rows; unpublished datasets do not count; the datasets need at least 80 % diversity. The LLM gateway is free and independent of credits (15M tokens per day, 2.5M per 5 minutes, 32 concurrent requests). GET /billing/balance returns your balance.
여기까지가 README 앞부분입니다. 나머지는 원문에서 읽을 수 있습니다. README 전문 보기 ↗
