Built-in agents
A fresh Exolvra install isn't empty. Sixteen specialist agents and four chatbot templates are seeded on first launch, covering the common roles you'd otherwise spend a day writing. This page is the catalogue — what each one is good at, what tools it expects, and when to clone rather than use as-is.
How the roster works
On first launch, Exolvra seeds sixteen specialist agents and four chatbot templates into the instance. You see them at /agents and /bots respectively, ready to use without configuring anything. If a built-in is missing at startup — typically because you deleted it, or because you’re upgrading from a version that didn’t have it yet — Exolvra seeds it back in with the default shape. Edits you made to existing built-ins are preserved across restarts.
This leads to three patterns of use:
- Use as-is — the default description, personality, and tools fit your case. Just assign work to it.
- Edit in place — tweak the personality, rename, change the icon, adjust the tool set. Your edits persist; restarts don’t overwrite them.
- Clone and customize — start from a built-in, clone it, then edit the clone freely. The clone is independent from the built-in: deleting the original doesn’t affect your clone, and re-seeding the original doesn’t touch your clone.
All specialists are autonomous — they have a heartbeat and can pick up assigned issues. All four chatbot templates are interactive-only — they live in /bots and respond to user messages. See Agents & chatbots for that distinction.
The specialists
The first nine cover the core professional roles and are documented in full below. A further seven — Chief of Staff and its reports, plus the Designer, App Builder, and Wiki Editor — round out the roster (see More specialists).
🤖 Exolvra Assistant (default)
General-purpose assistant that can chat, use tools, follow skills, and handle any task.
When to use. The default fallback. If a user chats without picking a specific agent, they’re talking to this one. Good for ad-hoc questions, quick tool-using tasks, and anything that doesn’t clearly belong to a specialist.
Heartbeat. 5 minutes.
Tool suggestion. Broad — web search, memory, data store, file store, think. Grant whatever the instance can afford without a specialty focus.
When to clone instead. If you want a “team assistant” with a specific voice or domain knowledge, clone this and edit the personality. The default is deliberately neutral; your variant can be opinionated.
📋 Project Manager
Autonomous project orchestrator. Evaluates project goals, creates and assigns issues to specialist agents, tracks progress, and adapts plans based on completed work.
When to use. Every project with more than two goals benefits from assigning the PM. Hand it a project with a clear goal and walk away — it decomposes the goal into issues, picks the best-fit specialist for each, links dependencies, and tracks progress on its heartbeat.
Heartbeat. 3 minutes — more active than most specialists because it’s coordinating.
Tool suggestion. project, task_board, think, architect, sessions_send, memory, send_email. The PM is a coordinator, not a doer — it doesn’t need shell or file_system.
Special behavior. The PM can create new specialist agents on demand. If a goal needs a skill no existing agent covers, the PM calls project.create_agent to mint one. New PM-created agents automatically report to the PM as their manager.
When to clone instead. If you want a second PM for a different workstream — say, a “Marketing PM” and an “Engineering PM” — clone this, rename, and scope the personality to that domain. The default PM is generalist.
💻 Code Assistant
Software development specialist for writing, reviewing, and debugging code across languages and frameworks.
When to use. Any issue involving code — writing a feature, reviewing a PR, tracing a bug, scaffolding a new project. The personality is tuned to read the codebase first (via explore and architect) before writing anything, and to prefer readability over cleverness.
Heartbeat. 3 minutes.
Tool suggestion. explore, architect, file_system, shell, web_fetch, data_store, think. The explore tool is its primary lens on the code. Also compatible with CLI adapters (Claude Code, Codex) — see CLI adapters for when to prefer those over the API path.
When to clone instead. Language-specific variants (python-assistant, rust-assistant) benefit from a cloned personality that encodes idioms and tooling for that language. Start from Code Assistant, add language-specific rules.
🔍 Research Analyst
Deep research agent that searches the web, reads sources, cross-references facts, and delivers structured findings.
When to use. Any task where the answer lives on the open web — competitive analysis, technical deep-dives, product research, policy summaries. The personality is tuned to plan with think first, cross-reference at least two sources before claiming a fact, and cite source URLs.
Heartbeat. 3 minutes.
Tool suggestion. web_search, web_fetch, data_store, memory, think. Optionally document_search if your instance has a knowledge base with relevant docs.
When to clone instead. Domain-specific analysts (competitive-intel, scientific-lit, market-research) gain a lot from a specialized personality. Clone and tell the variant what sources to trust, what formats to output, and what quality bar to hit.
⚙️ DevOps Engineer
Infrastructure and DevOps specialist for Docker, Kubernetes, CI/CD, cloud platforms, and system automation.
When to use. Anything involving containers, orchestrators, IaC, or cloud CLIs. The personality defaults to running terraform plan before apply, warning before destructive ops, and preferring IaC over click-ops.
Heartbeat. 2 minutes — faster than most because infrastructure work benefits from quick feedback loops.
Tool suggestion. shell, file_system, web_fetch, data_store, think. Pair with docker, kubernetes, terraform, aws-cli, gcloud, azure-cli skills for the specific stacks you use.
When to clone instead. If you work in a single cloud (AWS only, GCP only), a clone with the off-cloud skills removed is faster and less distracting. Also consider a “staging-only” clone that refuses to touch production.
✍️ Creative Writer
Content creator for blog posts, documentation, marketing copy, emails, and creative writing.
When to use. Any long-form writing — blog posts, release notes, landing page copy, emails to customers. The personality is tuned to read prior research first (so it builds on what the Research Analyst produced rather than re-researching), match audience and intent, and vary sentence length for rhythm.
Heartbeat. 5 minutes.
Tool suggestion. data_store, memory, file_store, web_search (for fact-checking), think, send_email. The blog-writer and email skills are useful pairings.
When to clone instead. Brand-specific writers carry the brand voice in the personality — clone and paste your brand guidelines into the system prompt. One clone per brand/voice.
📊 Data Analyst
Data analysis specialist for SQL queries, data exploration, statistical analysis, and visualization.
When to use. Any question that requires touching data — SQL queries, pandas transformations, statistical summaries, chart generation. The personality is tuned to start with exploratory queries (row counts, nulls, distributions) before diving into analysis, and to flag data quality issues proactively.
Heartbeat. 5 minutes.
Tool suggestion. shell (for Python/pandas scripts), data_store, notebook (for .ipynb files), web_search, think. Pair with database skill for direct SQL execution and math for stats.
When to clone instead. If you work with a specific warehouse (Snowflake, BigQuery) and a specific dialect matters, a clone with that dialect baked into the personality is worth it. Also consider a clone for a specific domain (finance-analyst, growth-analyst) that knows your metrics vocabulary.
🖥️ System Administrator
System administration agent for server management, networking, monitoring, security auditing, and troubleshooting.
When to use. Server health checks, log investigation, firewall rule review, permission audits, service restarts. The personality is tuned to gather state before suggesting fixes, warn before destructive ops, and check obvious causes (disk, DNS, permissions) before complex theories.
Heartbeat. 2 minutes.
Tool suggestion. shell, file_system, web_fetch, data_store, think. Pair with ssh, system-health, healthcheck, tmux skills.
When to clone instead. Clone for specific server roles (web-admin, database-admin) that have narrower scope and different default commands. Also consider a read-only audit clone that can diagnose but not change.
🏠 Home & Life Assistant
Personal assistant for smart home control, scheduling, reminders, weather, music, and daily life management.
When to use. Personal / household use cases — calendar reminders, weather-based suggestions, music control, smart home scenes. The personality is warm and conversational, not the terse reference tone of the work specialists.
Heartbeat. 5 minutes.
Tool suggestion. scheduler, user_profile, memory, data_store. Pair with weather, hue, spotify, calendar, contacts, translate skills depending on which services you’ve wired up.
When to clone instead. Family members who share an instance can each have a clone tuned to their schedule, smart-home rooms, and preferences. Otherwise the default handles most households.
More specialists
These cover executive, go-to-market, and platform-native roles added since the original nine. They follow the same use / edit / clone model.
🧑💼 Chief of Staff
Executive support specialist. Keeps the founder or exec organized — daily briefings, inbox triage, meeting prep, and flagging what matters before it slips. Acts on the user’s behalf; not a project coordinator.
When to use. Personal executive support rather than project delivery. It manages a team of its own — Sales BDR, Recruiter, and Investment Analyst report to it. Pair it with a daily briefing schedule.
📈 Sales BDR
Inbound + outbound business development specialist. Watches the sales inbox for replies, classifies intent (positive / neutral / negative), drafts personalized responses in the team’s voice, and escalates when a human should take over.
When to use. Pipeline development and first-touch responses. Reports to the Chief of Staff. Pair with email and CRM integrations.
🧲 Recruiter
Technical and business hiring specialist. Sources candidates, evaluates fit against a rubric, drafts personalized outreach, and tracks a hiring pipeline.
When to use. Active hiring. Give it a role rubric and let it source and track. Reports to the Chief of Staff.
💰 Investment Analyst
Deal-evaluation specialist for investors and operators doing M&A / partnership / investment due diligence. Researches companies, scores them against a thesis, and produces investor memos.
When to use. Due-diligence and deal evaluation. Hand it a thesis and a target list. Reports to the Chief of Staff.
📚 Wiki Editor
Maintains each project’s wiki — agent-authored, human-readable docs rebuilt on a schedule from closed issues, data store entries, approved learnings, and goals.
When to use. Keeping a project’s knowledge base current automatically. It runs on a schedule; you rarely assign it directly.
🎨 Designer
Produces interactive HTML mockups for design issues, behind a tight tool allowlist (canvas,
design_write, the visual checks, plus research and memory). No shell, file system, or general browser automation.
When to use. Design exploration in Design Studio. Assign it design issues; it iterates on a canvas and runs its own visual checks.
🧩 App Builder
Builds internal single-page apps with a NoSQL data backend and Exolvra SSO. Owns the full app lifecycle (
app_create/write/check/publish/rollback/archive), behind a tight allowlist — no shell or file system. Apps land at/apps/{slug}.
When to use. Shipping Internal Apps from a goal. Assign it an app-building issue, or let the PM dispatch one.
The four chatbot templates
Chatbot templates are single-purpose interactive bots pre-configured with a personality and conversation starters. They live in /bots. All four are designed to be cloned — the seeded version is the starting point, and your clones get the Bot API keys, knowledge collections, and branding overrides that make them yours.
🛠️ Support Assistant
Friendly customer support bot that answers questions, troubleshoots issues, and knows when to escalate.
Personality. Four goals in order: answer the question, troubleshoot, ask one clarifying question if unclear, escalate to human when stuck. Never invents policies or features.
Conversation starters:
- “I’m having trouble logging in”
- “How do I reset my password?”
- “Where can I find my billing history?”
- “Something isn’t working — can you help?”
Typical pairings. Attach a knowledge collection of your product docs, help-center articles, and policy pages so the bot can cite correct answers. See Memory & knowledge for how collections feed RAG.
When to pick it. Public-facing support widget on a product website. Escalation path to email or a human agent for things the bot can’t answer.
💼 Sales Q&A
Pre-sales assistant that explains products, compares plans, and helps visitors pick the right option.
Personality. Asks about the visitor’s use case before recommending anything. Honest about trade-offs, no overselling. Offers to connect to a human sales rep when the visitor is ready to buy or needs custom terms.
Conversation starters:
- “What does your product do?”
- “Which plan is right for me?”
- “Do you offer a free trial?”
- “Can I talk to a sales rep?”
Typical pairings. Attach a knowledge collection of product pages, pricing tiers, comparison matrices, and customer case studies. The bot will cite these rather than guessing.
When to pick it. Pre-sales widget on a landing page or pricing page. Pair with a form or booking tool for the “connect with sales” hand-off.
🚀 Onboarding Guide
Walks new users through first-time setup, explains key concepts, and celebrates their first wins.
Personality. Gets users from zero to first success as quickly as possible. Breaks setup into small steps, confirms each before moving on, adapts to the user’s pace. Celebrates the first success briefly before asking what’s next.
Conversation starters:
- “I just signed up — where do I start?”
- “How do I create my first project?”
- “What should I do first?”
- “Give me the 2-minute tour”
Typical pairings. Attach a knowledge collection of your getting-started docs, quickstart guides, and first-run checklists. Optionally wire the bot’s widget to inject the new user’s id and progress state so the guidance can be tailored.
When to pick it. Post-signup in-app widget for the first N days of a new user’s life. Swap to Support Assistant once the user is past onboarding.
📚 FAQ Bot
Answers common questions from a document collection. Pair with a knowledge base for grounded responses.
Personality. Answers only from provided knowledge. Says “I don’t have that information” rather than guessing. Short answers (2–3 sentences default), cites source documents when relevant.
Conversation starters:
- “What is this product?”
- “How does pricing work?”
- “Do you have a refund policy?”
- “What integrations do you support?”
Typical pairings. This is the template most dependent on a knowledge collection — without one, it can only say “I don’t have that information.” Attach a collection of FAQ docs, policies, and product pages.
When to pick it. Any surface where users ask repetitive factual questions and you want strict grounding. Documentation sites, help centers, internal intranets.
Clone, edit, or leave alone
| If you want to… | Do this |
|---|---|
| Use the built-in exactly as shipped | Assign it work. Done. |
| Change tool set, model, or heartbeat | Edit in place from the agent detail page |
| Rewrite personality, rename, or change the icon | Edit in place. Your changes persist across restarts. |
| Keep the built-in and have a customized version | Clone. The clone is independent of the original — your edits are isolated from any future changes to the built-in definition. |
| Delete a built-in | You can. It comes back on the next restart with the default shape. Disable instead if you want it out of the gallery without re-seeding. |
The three-pattern model — use / edit / clone — is the same for specialists and chatbot templates.
Where to go next
- Agents & chatbots — the conceptual difference between specialists and chatbots
- Your first agent — step-by-step for creating your own specialist
- Agents dashboard — where you edit, clone, and manage specialists
- Bots dashboard — where you edit, clone, and manage chatbots
- Bot test configurations — six additional bot recipes you can paste into the creator
- Orchestration patterns — how to combine specialists into a coordinated team