Eldric ships a family of specialized AI agent types. Each one is built for a specific kind of work and carries a carefully chosen set of tools — nothing more. That focus pays off three ways: safety, because an agent can't step outside its role; efficiency, because a narrow tool set means faster, sharper responses; and predictability, because behaviour stays consistent and easy to reason about.
A general-purpose assistant with every capability at once is hard to trust and hard to predict. Eldric takes the opposite approach: pick the agent that matches the task, and it brings exactly the tools that task needs. Read-only agents stay read-only. Code agents write code but never run shell commands. Network agents reach remote machines but never touch your local files. You always know what an agent can — and can't — do before you ask it anything.
Each card lists what the agent is for. Switch between them as your task changes — explore a codebase read-only, hand off to a code agent, then run the result.
The all-purpose default with full tool access. Reach for it when a task spans several domains, or when you're not yet sure which specialist fits.
Deep knowledge search that follows references across your knowledge bases and the web to build a well-sourced answer.
Reads, writes, and edits code. Implements features, fixes bugs, and refactors — without running shell commands, so changes stay safe.
Fact-checks claims and verifies sources. Use it to confirm an answer holds up before you act on it.
Read-only analysis and planning. Breaks complex work into a step-by-step implementation plan grounded in your actual code.
Compares options, summarizes material, and distils findings into clear takeaways for decision-making.
Safe, read-only codebase navigation. Searches, reads, and traces patterns — perfect for onboarding to unfamiliar code with zero risk of change.
Runs commands and scripts — tests, builds, deployments, diagnostics. Reads files for context but doesn't edit them directly.
Web research specialist. Searches, fetches documentation, and gathers external information — without touching local files or commands.
Connects to databases, explores schemas, and runs queries across PostgreSQL, MySQL, SQLite, and more. Read queries are safe; writes ask first.
Builds and queries knowledge bases with retrieval-augmented generation. Your AI grows smarter about your domain — no model retraining.
Network diagnostics and remote operations. Checks connectivity, traces paths, and manages remote machines for DevOps and infrastructure work.
Web crawling and structured data extraction. Walks sites and pulls the information you need into a usable shape.
Email and calendar management. Reads, searches, and sends messages, and keeps your schedule in view.
Infrastructure automation. Combines remote access, command execution, and file editing to provision and configure systems.
Real work flows across agents. You might explore a codebase read-only, plan the change, hand it to a code agent, then run the tests — each step in the hands of the specialist built for it. Because every agent's tools are scoped up front, you can chain them with confidence: the explorer never edits, the planner never executes, and the coder never runs commands it wasn't asked to.
In the chat shell and CLI you switch agents on the fly, list the tools available to the current agent, and check which agent and model are active at any time.
Beyond the built-in types, Eldric gives you two complementary ways to create agents tailored to your work.
Eldric's "AI that builds AI." Describe what you need in plain language and the Builder reasons through an agent design, drafts its tools, tests it against sample scenarios, and prepares it for deployment to your cluster. Best for novel, one-off agents where you want creative design rather than a fixed mould.
A fast, deterministic factory. Pick a domain template — finance, medical, DevOps, legal, science, support — and the Generator produces a consistent, production-ready agent with domain-appropriate tools, knowledge-base bindings, and safety guardrails baked in. Best for clear domains where you want repeatable results at speed.
Both run on your own cluster, with no data leaving your network. Building agents programmatically? See for developers.
Install Eldric on your own hardware and the full agent family is ready out of the box — explore, plan, build, run, research, and manage, each with the right tool for the job. Pick the agent, ask your question, and switch whenever the task changes.