pulsehive-core 1.0.1

Core traits and types for PulseHive multi-agent SDK
Documentation

pulsehive-core

Core traits and types for the PulseHive multi-agent SDK.

This crate defines the public API surface — all traits, type definitions, and error types used across the PulseHive ecosystem. It has zero runtime dependencies on LLM providers or HTTP clients.

Most users should use the pulsehive meta-crate instead of depending on this crate directly. Use pulsehive-core only if you're implementing a custom LLM provider or embedding provider.

Core Traits

Trait Purpose
LlmProvider LLM chat + streaming interface
Tool Pluggable agent capability (name, description, parameters, execute)
EmbeddingProvider Domain-specific embedding models
ApprovalHandler Human-in-the-loop approval for tool execution
ExperienceExtractor Custom experience extraction from conversations

Key Types

Type Description
AgentDefinition Agent blueprint (name + kind)
AgentKind Llm, Sequential, Parallel, or Loop
Lens Perception filter (domains, type weights, recency curve, attention budget)
LlmConfig Provider + model + temperature + max_tokens
HiveEvent 14 lifecycle event variants for observability
ToolResult Text, Json, or Error result from tool execution
PulseHiveError Top-level error enum (Substrate, Llm, Tool, Agent, Config, Validation, Embedding)

Implementing a Custom Provider

use async_trait::async_trait;
use pulsehive_core::llm::*;
use pulsehive_core::error::Result;

struct MyProvider;

#[async_trait]
impl LlmProvider for MyProvider {
    async fn chat(
        &self,
        messages: Vec<Message>,
        tools: Vec<ToolDefinition>,
        config: &LlmConfig,
    ) -> Result<LlmResponse> {
        // Your LLM API call here
        todo!()
    }

    async fn chat_stream(
        &self,
        messages: Vec<Message>,
        tools: Vec<ToolDefinition>,
        config: &LlmConfig,
    ) -> Result<Pin<Box<dyn Stream<Item = Result<LlmChunk>> + Send>>> {
        todo!()
    }
}

Links

License

AGPL-3.0-only