NexusPilot

NexusPilot is a multi-database workbench for developers and data teams. Its agent works from real connection state, engine-native objects, and data results to explore different data models in natural language and act through controlled workbench tools. It is built on Tauri v2, React 19, and Rust, with a local AI Runtime sidecar powered by Bun + Elysia + the Vercel AI SDK.

✨ Core Capabilities

- Natural-language data agent — use Ask, Query, and Agent modes to describe data questions or multi-step tasks in natural language. The agent works from real connection state, engine-native objects, and data results to query, explore, explain, and operate on data according to each source's capabilities.

- Controlled agent tool execution — the agent uses data sources through restricted workbench tools instead of opening hidden connections outside the application. Every operation is checked against driver capabilities, resource type, and risk level; SQL and Redis writes currently use risk analysis and approval flows, with strong confirmation for high-risk actions.

- Multi-source connections and object exploration — manage relational, key-value, and analytics sources in one connection tree, with more data models designed to fit the same capability system. Each engine exposes its own object hierarchy and available operations instead of being forced into one table abstraction.

- Engine-aware query and operation workspaces — use interaction surfaces adapted to each source. Current workflows include context-aware query editing, multi-statement results, saved queries, and Redis Key, typed-value, and TTL operations; new sources can add native workflows without pretending to be SQL or tables.

- Data inspection and safe mutations — inspect content through views appropriate to the current resource. Reading, filtering, paging, and mutation entry points follow source and resource capabilities; supported writes use previews, stable identity, preconditions, or transactions to reduce accidental and concurrent overwrites.

- Engine-native object management — inspect and manage the objects and structures each source actually owns. Current relational drivers provide table inspection, structured design, and DDL previews, while ClickHouse adds native workflows for databases, tables, views, projections, and data-skipping indexes.

- Bring-your-own models with a local AI Runtime — manage providers, models, and credentials in the local AI Runtime, synchronize the models.dev catalog, and configure custom OpenAI-compatible providers. The agent reuses workbench connections through a controlled bridge; the frontend never stores LLM credentials, and the workbench remains usable without AI.

- End-to-end encrypted device sync — use NexusPilot Cloud to synchronize saved connections and folders across authorized devices, with device authorization, conflict handling, and Recovery Key recovery. Cloud stores ciphertext rather than plaintext credentials, while the local workbench and AI Runtime remain independently usable.

🗄️ Supported Databases

| Database | Data model | AI tool integration | Status |

| ---------- | ------------------ | ------------------- | ----------- |

| PostgreSQL | Relational | ✅ Integrated | ✅ Supported |

| MySQL | Relational | ✅ Integrated | ✅ Supported |

| SQLite | Relational | ✅ Integrated | ✅ Supported |

| Redis | Key-value | ✅ Integrated | ✅ Supported |

| ClickHouse | Columnar analytics | ✅ Integrated | ✅ Supported |

| Oracle | Relational | ✅ Integrated | ✅ Supported |

This matrix describes the public availability of connection drivers; it does not imply identical operations across database engines. “AI tool integration” means the AI Runtime can use the driver’s currently registered connection, object, or data tools through the workbench connection runtime—not that every read, write, or management operation is enabled. Exact capabilities depend on the engine, server version, connection permissions, and runtime capability model.