The Vadalog API covers program analysis, bind annotation management, and direct evaluation of Vadalog programs against the workspace-aware engine. All paths are relative to the base URL and require authentication. All responses use the standard envelope.
These are the platform Vadalog endpoints — they are authenticated, workspace-aware, and route to the correct compute target for the authenticated user. For the raw low-level engine API (unauthenticated, direct engine calls), see Engine API.

Analyze a program

Inspect a Vadalog program’s predicate structure without executing it. Handles Vadalog, SQL, Cypher, and Python concept types.

Example

Validate a concept

Check a concept body and return a verdict, without saving anything and without running the engine. Use it before saving a concept to surface errors early. What gets checked depends on concept_type: logic compiles the Vadalog program, sql and cypher are converted to a Vadalog rule, python is parsed for syntax errors, and llm and context — which have no compilable body — have their concept_config checked structurally instead.
This differs from /vadalog/analyze, which reports a program’s predicate structure rather than a pass/fail verdict.

Example

data.valid is always present, so a caller gets an explicit verdict. Note that a failed validation is still an HTTP 200 with status: "success" — read data.valid, not the status code:

Parse binds

Parse @bind and @qbind annotations from a Vadalog program and separate the rule code from the datasource bindings.

Example

Evaluate a program

The primary platform execution endpoint. Evaluate a Vadalog program with optional runtime parameters and compute target. The platform resolves the correct engine (local or remote compute) for the authenticated user.

Example

For concept-level execution (with dependency chaining, result persistence, and progress tracking), use POST /concepts/{ontology_id}/run/{concept_name} instead. Use /vadalog/evaluate when you want to execute a program directly without a saved concept.

Generate a program from a question

Write a Vadalog program from a natural-language question. Nothing is evaluated or saved: pass the program to /vadalog/evaluate to run it. Each successful call is counted in your usage as one program generation.

Example

The response is the translator’s answer as is, not the standard envelope. code_examples holds the program, and response the full answer in markdown:
An error from the translator keeps its own HTTP status. A missing query returns 400, and a translator that cannot be reached returns 503.

Build a bind

Rewrite an existing @bind or @qbind annotation for a new predicate name. When isOutput is true, the output target (CSV filename or query name) is also updated to match the new predicate.

Example

Search the Vadalog docs

Search the Vadalog documentation and return a ready-to-use code example for the query. It answers with code, not prose — it is built for a coding assistant that needs a worked example of a language feature, and it backs the search_vadalog_docs MCP tool.
data.vadalog_code is the best-matching example; total_examples says how many were found. A query that matches nothing returns 404 rather than an empty example.

Engine status

Return whether the Vadalog engine is idle or processing an evaluation. Use this to gate new submissions or display engine state in a UI.
No parameters.

Stop evaluation

Cancel the current evaluation. Issues a stop signal to both the Vadalog engine and any in-flight Python concept subprocess, so a single call covers both concept types.
No parameters.
The response body is whatever the engine returns on acknowledgement of the stop request.
The /vadalog/spark-ui/* and /vadalog/actuator/* proxy endpoints exist for operational monitoring of the underlying engine (Spark job tracking, JVM metrics, log tailing). They are outside the scope of external integration and are not documented here.POST /vadalog/evaluateWithParams is likewise a transparent proxy: it takes the engine’s own form-encoded signature and returns the engine’s raw response, so that a client written against the Engine API can be pointed at the platform without changing. New integrations should use /vadalog/evaluate, which takes JSON and returns the standard envelope.