BotHive
A hosted platform for defining, running, and operating AI agents with tools, memory, integrations, and durable runs.
The Story
BotHive is a product surface for creating and operating AI agents. The codebase connects hosted bots to runtime policies, model selection, tool descriptors, memory retrieval, execution records, and channel-facing APIs.
What I Built
The hard part is not putting a chat box around a model. A useful agent runtime has to decide what kind of work a request represents, load the right context, expose only appropriate tools, and leave enough state behind to inspect or resume a run. A zero-network classifier routes routine conversation, retrieval, memory, tool-use, reasoning, and HiveLang-backed requests to fast, smart, or agent tiers. The execution route layers access checks, rate and credit guards, context loading, memory recall, durable checkpoints, and audit events around a bot run. The implementation explicitly handles model-routing decisions, workspace and API-key access, verified-memory fallbacks, durable run transitions, and runtime status changes. Those boundaries make the product more legible than an opaque prompt-to-response pipeline.