/query Content-Type: application/json { "content": "" } ``` - **Request Body** - `content` (string) – raw text whose embedding is used for similarity search (currently converted to dummy zero‐vector). - **Responses** - `200 OK` + JSON array of matching IDs (`Vec`) ordered by descending similarity (highest score first). - `500 Internal Server Error` + plain text error message if ChromaDB query fails. - **Example** ```bash curl -i -X POST http://localhost:8080/api/mem/latent/query \ -H "Content-Type: application/json" \ -d '{"content":"search terms"}' ``` ### 5. Chat Endpoint ``` POST /api/chat Content-Type: application/json { "message": "" } ``` - **Request Body** - `message` (string) – user’s chat input. - **Behavior** 1. If a **Sentience DSL** is loaded, constructs DSL code snippet: ``` on input() { … } ``` – Calls `SentienceWrapper.handle_code(...)` synchronously. – After execution, checks if DSL wrote a `"response"` key in its internal short memory. If so, returns that as the response. 2. If DSL didn’t produce a response (or DSL isn’t loaded): - Checks for `remember `, `recall `, or `if context includes ` commands for BaseAgent logic. - Otherwise, falls back to LLM: calls `model::generate(message).await`, returning the generated text. - **Response** - `200 OK` + JSON string with the agent’s reply. - **Example** ```bash # Remember example curl -i -X POST http://localhost:8080/api/chat \ -H "Content-Type: application/json" \ -d '{"message":"remember mood = happy"}' # Recall example curl -i -X POST http://localhost:8080/api/chat \ -H "Content-Type: application/json" \ -d '{"message":"recall mood"}' # Generic question (LLM fallback) curl -i -X POST http://localhost:8080/api/chat \ -H "Content-Type: application/json" \ -d '{"message":"Tell me a joke."}' ``` ### 6. Sentience DSL Endpoint ``` POST /api/sentience/run Content-Type: application/json { "code": "" } ``` - **Request Body** - `code` (string) – raw Sentience DSL snippet to execute immediately (without `on input(...)`). - **Behavior** - Wraps the provided `code` string and calls `SentienceWrapper.handle_code(code)`. - Returns whatever the DSL code returned (if any) or an empty string if nothing was returned. - **Response** - `200 OK` + JSON: ```json { "output": "" } ``` - `output` may be empty (`""`) if DSL script didn’t explicitly set a value. - **Example** ```bash curl -i -X POST http://localhost:8080/api/sentience/run \ -H "Content-Type: application/json" \ -d '{"code":"on startup() { mem.short[\\"greeting\\"] = \\"Hello from DSL!\\" }"}' ``` --- ## Agent DSL (“Sentience”) Details **Sentience DSL** allows you to define event‐driven logic in a simple, internal scripting language. Typical patterns: 1. **Loading DSL at Startup** - Place your DSL file as `agent.sent` in project root or specify via environment variable. - On server startup, `main.rs` does: ```rust if let Ok(sent_code) = fs::read_to_string("agent.sent") { base_agent.load_sentience(&sent_code)?; println!("Loaded Sentience DSL from agent.sent"); } ``` 2. **DSL Grammar (Example)** ```sentience // Defines behavior on input event on input(user_msg) { if user_msg == "ping" { mem.short["response"] = "pong from DSL"; } else if user_msg startswith "set " { // parse "set key=value" let kv = split(user_msg[4..], "="); mem.short[kv[0]] = kv[1]; } } // Defines behavior on startup on startup() { mem.short["welcome"] = "Welcome, dear user!"; } ``` 3. **Memory Access in DSL** - `mem.short[] = ` reads/writes short-term memory. - `mem.long[] = ` reads/writes long-term memory (SQLite). - The wrapper seeds both memories at `handle_code` entry and flushes any writ

Inception ICORE Server

工具