Agents

Agents are the autonomous workers in Cendriix. Each agent has a role, model, permitted tools, and resource budget. Access the agent library via Automations → Agents library (G A).

Agent Library

The Agent Library is your discovery and management hub for all agents — built-in and custom.

Tabs

Tab	Description
Browse	Full catalog of available agents (built-in + community) with filter rail
Mine	Agents you’ve created or forked
Recipes	Pre-built agent configurations for common use cases
Build	Wizard to create a new agent from scratch

Agent Card

Each agent card displays:

Agent name and icon
Trust pill (verified, community, custom)
One-line description
Permitted tools count
Blast-radius badge (low / medium / high)
Hover actions: Fork, Configure, Run

Built-in agents

Agent	Role	Default model
Orchestrator	Plans, delegates, sequences	claude-opus-4
Dev	Code generation, PRs, CI	claude-sonnet-4
SRE	Infrastructure, rollbacks, incidents	claude-sonnet-4
QA	Test generation, validation	claude-sonnet-4
Security	Policy enforcement, secret scanning	claude-sonnet-4

Building an agent

Via the UI wizard

Navigate to Agents library → Build (or click ”+ Build” in the header).
Name your agent, choose a base model, and write the system prompt.
Select permitted tools from connected integrations.
Set resource budget (max tokens, max cost USD).
Optionally configure a Cendra recommendation prompt to suggest this agent for matching intents.
Publish — the agent is immediately available in flows and the Studio.

Via YAML definition


slug: my-custom-agent
name: My Custom Agent
model: claude-sonnet-4
system_prompt: |
  You are a specialized agent for ...
permitted_tools:
  - aws.ec2.describeInstances
  - github.pulls.create
resource_budget:
  max_tokens: 100000
  max_cost_usd: 10.00

Deploy with the CLI:


cendriix agent push --file agent.yaml

Agent Studio

The Agent Studio (/agents/studio) is a dedicated testing and debugging environment. Run your agent against sample inputs, inspect the reasoning trail (tool calls, LLM turns, stop reasons), and iterate on the system prompt before publishing.

Forking

Fork any built-in or community agent to customize it. Forked agents retain versioned lineage back to the original, enabling upgrade pull when the upstream agent improves.


POST /entities/:id/fork

Versioning

Agents are versioned entities. Each publish creates an immutable snapshot:


POST /entities/:id/publish
GET /entities/:id/versions

In-flight runs use their pinned version; new runs pick the latest.

API

Method	Endpoint	Description
`GET`	`/agents`	List workspace agents
`POST`	`/entities`	Create agent (`kind: "skill"` with agent spec)
`GET`	`/entities/:id`	Get agent detail
`POST`	`/agent/steps`	Execute a single agent step
`GET`	`/agent/runs/:runId/steps`	List steps for a run

See also: Flows, Skills, Connectors, The Studio