Three APIs, one developer surface. Image Diagnosis (crop disease from a photo with AI focus areas). Machine Learning (crop recommendation and yield prediction). Text & Audio (translation and transcription). Each works inside agriculture by default, and a lot of them work outside it. This page introduces all three and the kinds of products we have seen developers build on top.
# Good API product design asks:
who: "farmer, advisor, extension team, researcher"
input: "image, text, location, weather, language"
risk: "low, review-needed, regulated decision"
output: "ranked signal + evidence + next-step guidance"
handoff: "show uncertainty, cite sources, escalate when needed"The Four APIs
Each API does one thing well and shares the same auth, billing meter, and locale handling as the others. Brief introductions below; the detailed developer reference for each is one click away.
Send a crop photo, receive a disease prediction with a confidence score and an AI focus-area overlay that shows what the model attended to. Four crops today: maize, rice, tomato, cassava. Stage the image with a presigned upload, then run raw-key inference; the same path serves both user-facing apps and server pipelines.
A content-agnostic translation and transcription service. Submit text or audio in your input language, choose a target language and an output format (text or audio), and, when the language pair is supported, receive the translated result back. Focused on local dialect coverage.
Crop recommendation from soil + climate inputs (Random Forest over 22 crops). Maize yield prediction from fertilizer + soil + climate (XGBoost on Sub-Saharan Africa trials). Both with reasoning traces, no black boxes.
Start Building
All of the APIs share one base URL, one auth header, and one credit meter. Here is everything you need to start.
Base URL https://api.fildraai.com/api/v1. Send your key on every request; either header works:
X-Api-Key: YOUR_API_KEY
Authorization: Bearer YOUR_API_KEY
User key: for your app's per-user inference calls.
Server key: for trusted backends; required for audio job polling and locale listing.
| Plan | Monthly allowance | Rate limit | Price |
|---|---|---|---|
| Free | 25 calls | 1 request / second | $0 |
| Basic | 150 credits | 5 requests / second | $15 / month |
| Pro | 600 credits | 20 requests / second | $50 / month |
| Call | Credits |
|---|---|
| Image diagnosis (infer) | 3 |
| ML crop, yield, or snapshot | 1 |
| Text translation | 1 (2 with audio output) |
| Audio job | Metered: base 1, plus 1 per 30s in and 1 per 30s spoken out |
| Knowledge reads and /locales | 0 (free) |
| Status | What it means and what to do |
|---|---|
400 | Validation failed: a missing or invalid field, or an unsupported value. Check the request against the docs. |
401 | Missing or invalid API key. Check the key and the header. |
403 | The key lacks the scope or audience this route needs. |
413 | Input over the size or length cap, for example text beyond your plan limit. |
429 | Rate limit exceeded. Back off and retry. |
5xx | Server error. Retry with backoff; contact us if it persists. |
Async jobs (audio, and text jobs) finish by polling; webhooks are not supported yet. There is no SDK to install: any HTTP client works, such as cURL, Python requests, or fetch.
Beyond Agriculture
We built these APIs for agriculture because that is where our research and field validation lives. The underlying surfaces, annotated images, curated text, a tag vocabulary, a translation pipeline that handles African languages, fit a lot of adjacent problems. Some ideas we have seen, some we expect:
Wherever a photo needs a first-pass classification with a confidence score, the same upload + inference pattern transfers. Quality-control screens on a packing line, citizen-science contributions, supply-chain inspection, pair our crop classes with the operator's manual review.
The reasoning-trace contract, input → ranked output + explanation, fits decision-support work outside agronomy too. Students learning ML can study a working RAG + classification system; researchers can use it as a baseline to benchmark against; product teams can prototype before training their own model.
We do not advertise these as primary use cases, agriculture is. But the API contract is general enough that imagination is genuinely the limit. If you build something interesting outside our home territory, tell us about it.
Developer Scenarios
These patterns show where FildraAI APIs fit inside real products. The endpoint documentation explains the exact request and response details.
Add image-based crop issue triage to a mobile app so farmers can capture a symptom, see ranked possibilities, and understand when expert review is needed.
Help agronomists, call centers, and extension officers review submitted images with consistent visual evidence, source-backed context, and clear uncertainty.
Support structured scouting, field trials, and regional monitoring by turning repeated submissions into comparable signals that can be reviewed over time.
Build multilingual agricultural experiences where the same core signal can be presented with local language, crop, region, and review context.
Pair model signals with practical context so users see what the output means, what it does not prove, and what should be verified locally.
Best Practices
Agricultural APIs are most useful when the surrounding product sets expectations, shows evidence, and keeps human judgment in the workflow.
Present ranked possibilities, confidence, and evidence clearly. Do not frame model output as a confirmed diagnosis or a pesticide instruction.
Show the image, top predictions, and visual focus areas together so users can inspect whether the output matches what they see in the field.
Ask for crop, location, growth stage, and recent field conditions before turning a model result into user-facing guidance.
Route severe, uncertain, regulated, or high-value cases to a qualified person instead of encouraging automatic action.
What Developers Get
The current solution is designed around authenticated HTTP endpoints, staged-image processing, and diagnosis support flows rather than a packaged SDK.
Current image diagnosis support covers cassava, maize, rice, and tomato under the models you have already integrated.
End users upload files to the API gateway, and the server handles storage transfer on their behalf to reduce client-side storage coupling.
Internal and server-side workflows can call the diagnosis API directly with a staged object key instead of re-uploading the image payload.
Diagnosis workflows are designed to pair ranked predictions with visual evidence so model outputs are easier to inspect and explain.
Knowledge-backed diagnosis endpoints can complement model predictions with structured profile content for interpretation and follow-up guidance.
Integration is done through documented HTTP endpoints and API keys. We do not currently advertise a dedicated SDK on this page.
Request developer access and we can point you to the right integration path across image diagnosis, machine learning, and text and audio.