> ## Documentation Index
> Fetch the complete documentation index at: https://docs.deepmyst.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Agent Builder
> Build AI agents in human language and ship them to every department
Agents are first-class processes on DeepMyst. You describe what you want in plain
language; DeepMyst picks the right frontend, attaches the right knowledge, and
applies your policy. The result is a working agent with a shareable URL — built in
a sentence, reliable in production, and audited forever.
## Build an agent in a sentence
Open the **Agent Builder**, describe the job, and click **Build agent**. DeepMyst:
1. **Reads your request** to understand the task and the people who'll use it
2. **Picks the right frontend** — chat, voice, document, workflow, and more
3. **Wires in knowledge** — handbooks, documents, datasources, and tools the agent
needs
4. **Applies your policy** — the same governance every request inherits
For example:
> *"As I type a vendor contract, review each clause against our procurement policy,
> flag risky language, and propose redlines."*
becomes a **Document** agent that reviews clauses in real time, cites your
procurement policy on every flag, and proposes redlines as suggestions — never
auto-applying them.
You can keep refining in the full builder: attach more knowledge, change the
frontend, set policy, and choose how the agent is published.
## Agent types
Every agent gets the interface that fits the work. DeepMyst ships a catalog of
frontends, each a first-class agent shape:
| Frontend | Best for |
| ------------ | -------------------------------------------------------- |
| Chat | Q\&A, copilots, onboarding buddies |
| Voice | Real-time call coaching and hands-free assistants |
| Document | Track-changes review, drafting, structured briefs |
| Workflow | Multi-step, scheduled, or triggered automations |
| Kanban | Triage and classification onto a board |
| Form | Structured intake that drafts an artifact at the end |
| Wizard | Multi-step vetting and guided processes |
| Code Editor | Diff review against a style guide |
| Spreadsheet | Working models that watch assumptions and flag anomalies |
| Dashboard | Quick actions and briefs on warehouse data |
| Image Studio | Posters and social posts in your brand voice |
| Presentation | Slide decks from a one-line brief |
| Custom App | Bespoke UI for memos, board decks, and OKRs |
## What every agent gets
No matter the frontend, every agent runs on the same stack and inherits the same
capabilities:
A knowledge graph, documents, connected datasources, and run history. The
persistent layer every agent reads from.
Built-in tools, MCP servers, web search, and sub-agents. Bring your own or use
the library.
Any of 500+ models behind one API, with optimization, routing, and automatic
failover.
Identity, cascading policy, and inline security applied to every run — never
bolted on after.
See [the OS stack](/architecture) for how these layers fit together.
## Every department, every workflow
A single platform produces the right agent for any team:
**Bug Triage** (Kanban) auto-classifies alerts onto a severity board.
**PR Reviewer** (Code Editor) reviews diffs against your style guide.
**Postmortem Intake** (Form) drafts the postmortem from structured input.
**Sales Coach** (Voice) coaches calls in real time and logs to your CRM.
**Brand Studio** (Image Studio) creates posts in your brand voice.
**Pitch Deck** (Presentation) builds decks from a one-line brief.
**Contract Reviewer** (Document) redlines against your procurement policy.
**Onboarding Buddy** (Chat) points new hires to tools, docs, and people.
**Vendor Intake** (Wizard) vets vendors and drafts the DPA.
**Metrics Brief** (Dashboard) acts on warehouse data.
**Financial Model** (Spreadsheet) maintains a working model and flags anomalies.
**Strategy Copilot** (Custom App) handles memos, board decks, and OKRs.
## Production-grade by default
Agents built in a sentence still behave like production software:
* **Evaluated** — every output is scored against your rubric before it reaches anyone
* **RBAC-aware** — responses are scoped to the role of the person using the agent
* **Replayable** — every run is logged forever and can be replayed on demand
* **Observable** — inputs, outputs, latency, tokens, and scores are recorded per run
## Skills and tools
Agents do more than answer. Through skills, they take action:
* **Built-in tools** for common operations
* **MCP servers** — connect any Model Context Protocol server for tool use
* **Web search** for live information
* **Sub-agents** — agents that call other agents to break down complex work
* **Datasources** — query Postgres, ClickHouse, Snowflake, and more
## Deployment
When an agent is ready, publish it the way your team will actually use it:
* **Shareable URL** at `your-org.deepmyst.app/agents/...`
* **Embed** it into an existing product or internal tool
* **Public portal** for external-facing agents
* **Visibility scopes** — organization, department, specific members, or projects
## Build a whole company
**Company Builder** is in private preview for design partners. [Get in touch](mailto:Sharif@deepmyst.com?subject=DeepMyst%20Company%20Builder%20-%20design%20partner)
to join.
Instead of building one agent at a time, you can build an entire AI-native company
in one job. Pick an industry, answer five questions, and DeepMyst:
* **Interviews your founder** in human language
* **Designs departments and teams**, flat or hierarchical
* **Curates an agent catalog** with rationale and consolidation logic — consolidated,
not bloated
* **Builds everything atomically** — folders, tags, knowledge bases, and policies —
into a working workspace
## Next steps
Set the policy, security, and access rules every agent inherits
See how agents, memory, skills, models, and governance fit together