AI Agent

The Vadalog AI Agent is a streaming conversational interface that can autonomously plan and execute data workflows — generating Vadalog programs, running concepts, inspecting results, and iterating based on outcomes.

Chat

Send a message to the agent and receive a streaming NDJSON response. The agent maintains conversation history within a session, so follow-up messages have full context. HTTP Request:

POST /api/v1/agent/{ontology_id}/chat

Path Parameters:

Parameter	Type	Required	Description
`ontology_id`	string	Yes	The ID of the ontology to work within

Request Body:

Field	Type	Required	Description
`message`	string	Yes	The user’s message
`session_id`	string	No	Conversation session identifier. Default: `"default"`
`model`	string	No	Override the LLM model for this session

Example:

curl -X POST "https://api.prometheux.ai/jarvispy/my-org/my-user/api/v1/agent/my-project-id/chat" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_JWT_TOKEN" \
  -d '{
    "message": "Create a concept that finds all transitive suppliers and run it",
    "session_id": "session-1"
  }'

Response (NDJSON stream): The response is a stream of newline-delimited JSON objects. Each line has a type and data field:

Type	`data` shape	Description
`metadata`	`{ ontology_id, model, context_loaded }`	Session metadata — the first frame of a turn.
`context_loaded`	`{ count, notes: [...] }`	Which context notes were pulled in for this turn.
`thinking`	`{ chunk }`	A chunk of the agent’s reasoning stream. Emitted continuously while the agent plans and runs tools — this is what fills the wait between the question and the answer.
`tool_start`	`{ tool, input }`	Agent is invoking a tool.
`tool_result`	`{ tool, result }`	Result from a tool invocation.
`proposal`	`{ tool, concept_name, definition, description }`	Agent proposes a concept/app/schema change.
`content`	`{ chunk }`	A chunk of the agent’s answer text. Concatenate chunks in order.
`done`	`{ ontology_id }`	Stream complete.
`session_meta`	`{ session_id, title }`	Streamed after `done` on the first turn, carrying the auto-generated session title.
`error`	`{ error }`	An error occurred.

Render thinking chunks live (e.g. a collapsible “thinking…” panel) so users see progress. Without it the UI looks frozen while the agent works, then dumps the whole answer at once. Handle unknown type values by ignoring them — new event types may be added over time.

Example stream:

{"type": "metadata", "data": {"ontology_id": "my-ontology-id", "model": "claude-sonnet-4-6", "context_loaded": true}}
{"type": "thinking", "data": {"chunk": "The user wants transitive suppliers. Let me create a recursive concept..."}}
{"type": "tool_start", "data": {"tool": "create_concept", "input": {"definition": "..."}}}
{"type": "proposal", "data": {"tool": "create_concept", "concept_name": "transitive_suppliers", "definition": "...", "description": "..."}}
{"type": "tool_result", "data": {"tool": "create_concept", "result": {"id": "concept-uuid"}}}
{"type": "tool_start", "data": {"tool": "run_concept", "input": {"concept_name": "transitive_suppliers"}}}
{"type": "tool_result", "data": {"tool": "run_concept", "result": {"total_records": 42}}}
{"type": "content", "data": {"chunk": "I created and ran the transitive_suppliers concept. It derived 42 records..."}}
{"type": "done", "data": {"ontology_id": "my-ontology-id"}}
{"type": "session_meta", "data": {"session_id": "session-1", "title": "Transitive suppliers"}}

Read-only vs. tool-authorizing clients. Sent as shown above, the agent runs its tools autonomously and streams the result. If you include a mode field in the request body, the agent instead pauses and emits authorization_request events for tools that need approval — your client must then reply (with approved_calls / authorized_tools) to continue. For a simple question-and-answer client, omit mode.

On the free tier the agent is capped (default 150 messages/month). Over the cap the endpoint returns HTTP 402 with a JSON body { "code": "AGENT_MESSAGE_LIMIT_REACHED", "used", "limit", "resets_at" } instead of a stream. Handle it as a distinct “limit reached” state.

Reset Session

Clear the conversation history for a session. HTTP Request:

POST /api/v1/agent/{ontology_id}/reset

Path Parameters:

Parameter	Type	Required	Description
`ontology_id`	string	Yes	The ID of the ontology

Request Body:

Field	Type	Required	Description
`session_id`	string	No	Session to reset. Default: `"default"`

Example:

curl -X POST "https://api.prometheux.ai/jarvispy/my-org/my-user/api/v1/agent/my-project-id/reset" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_JWT_TOKEN" \
  -d '{"session_id": "session-1"}'

Session ManagementSessions are maintained server-side and keyed by username + session ID. Use different session_id values to run parallel conversations. Sessions are automatically evicted (LRU) when the server reaches capacity.

Introduction

Get Started

Platform

Integrations

Examples

Vadalog Reference

Resources

Chat

Reset Session

​Chat

​Reset Session

Chat

Reset Session