Zonic AIAPI Reference

Dekiru Editing And QA

Use Dekiru chat for grounded VideoQA over analyzed footage, and for building or revising a renderable edit sequence.

Choose An Interface

Dekiru is the editing and QA layer. It consumes Nandemo's scene, transcript, metadata, people, and search outputs, then uses tools or chat turns to answer higher-level questions and produce edit plans. Use the lowest-level interface that matches your product need.

  • Use /videos/search for direct semantic and label retrieval.
  • Use /mcp when an MCP client wants structured read-only retrieval tools over Nandemo outputs.
  • Use /chat for VideoQA and Dekiru editing workflows: answers, edit sequences, revisions, references, and render preparation.

Retrieval MCP

POST/api/projects/{pid}/mcp

The MCP endpoint is JSON-RPC 2.0. Call tools/list first if your client wants to discover tools dynamically.

MCP is best when your application wants structured retrieval without opening a full assistant conversation. The caller selects a tool, passes structured arguments, and receives a structured result. Conversational VideoQA should go through chat, where Dekiru can fetch evidence and synthesize the answer.

curl -X POST "$BASE_URL/projects/$PID/mcp" \
  "${AUTH[@]}" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'


curl -X POST "$BASE_URL/projects/$PID/mcp" \
  "${AUTH[@]}" \
  -d '{
    "jsonrpc": "2.0",
    "id": 2,
    "method": "tools/call",
    "params": {
      "name": "search_script",
      "arguments": { "query": "pricing and availability", "limit": 5 }
    }
  }'

Chat Stream

POST/api/projects/{pid}/chat

Chat returns newline-delimited JSON. Parse the stream line by line and handle events such as text, thinking, tool_output, plan, done, and error.

Chat is for user-facing QA and editing workflows. A user can ask Dekiru a grounded question about the footage, ask it to find material, reason about story structure, create a sequence, revise that sequence, or prepare a render. The stream is event-based so your UI can show partial assistant text, tool evidence, generated plans, and completion state without waiting for the full response.

curl -N -X POST "$BASE_URL/projects/$PID/chat" \
  "${AUTH[@]}" \
  -d '{"message":"Which shots show the product demo?"}'


curl -N -X POST "$BASE_URL/projects/$PID/chat" \
  "${AUTH[@]}" \
  -d '{"message":"Create a concise 30 second product launch recap"}'

Sequences And Render

GET/api/projects/{pid}/plans

Sequences are generated by Dekiru chat. A sequence is a structured edit plan: selected source clips, timing, ordering, and editorial intent. Retrieve plans when your integration needs to display generated edits, store a selected plan, or pass one to rendering.

PLAN=$(curl -s "$BASE_URL/projects/$PID/plans" \
  -H "x-api-key: $API_KEY" | jq '.plans[0]')
POST/api/projects/{pid}/render

Rendering turns a sequence plan into an exported video. The render endpoint expects the plan structure produced by Dekiru, so most integrations call it after chat has emitted a plan or after the user has chosen a saved plan.

curl -X POST "$BASE_URL/projects/$PID/render" \
  "${AUTH[@]}" \
  -d "{\"plan\":$PLAN}"

References

POST/api/projects/{pid}/references

Upload image or text references before a chat turn when the assistant should inspect external context in addition to the project footage.

References are project-scoped context for Dekiru. They can represent a creative brief, brand note, example image, or other material that should influence QA and sequence generation. They do not replace Nandemo analysis; they guide how Dekiru interprets and edits the already analyzed footage.

curl -X POST "$BASE_URL/projects/$PID/references" \
  "${AUTH[@]}" \
  -d '{"kind":"text","name":"brief.txt","mime":"text/plain","text":"Use an energetic launch tone."}'
Back: Retrieve understandingNext: MCP connector