loading
Loading content
Announcement
Describe the work. Agents build the graph.
They compose real tools into a DAG, place agent nodes beside them, and run across the fleet. MCP from Claude Code, Cursor, or Codex.
the graph
The graph is the methodology.
Agents compose tools into a DAG and sit on it as nodes. One orchestrated run across the fleet.
the fleet
The graph becomes a run.
Each node is a container. Agent nodes included. The fleet runs them in parallel.
live tables
Every output becomes a row.
Findings land structured and deduplicated. One row per finding, not a folder of logs.
app mode
Agents ship end-to-end apps.
Internal tools or customer-facing surfaces over live tables. App Mode builds them; the SDK drives the same stack.
workflow nodes
Agent nodes sit on the graph.
Drop an AI Agent node next to tools and scripts. One graph. One fleet run.
Workflowsthe fleet
The fleet grows around the run.
Each ready node takes one machine. More waiting nodes means more machines join.
ExecutionManaged
pool scales with the run
- 4 on jobs
- +2 joining
Self-hosted
hosts you enroll take jobs
- 2 on jobs
- +1 joining
live tables
Ask the table what changed.
One TQL line filters the live table. Each row marks new, resurfaced, or gone.
Databasesseverity = "critical" OR severity = "high"
- vpn.target.trickest.net
- api.target.trickest.net
- dev.target.trickest.net
app mode
End-to-end apps, not exports.
Internal dashboards or external portals on the live table. Agents build in App Mode; wire them in with the SDK.
SDKThe features that make the agent work
Here are the most important parts, one row per feature. Building and running first, then the agent side.
Console and editor
Build and run without the agent frame
canvas
One canvas for build and run.
Two nodes and one wire: the graph you edit is the graph that runs, with no Builder to Executions hop.
nodes
Click a node. Get its workspace.
Open any node to configure inputs, read the command, and run it without leaving the canvas.
integrations
Integration support.
Connectors are ordinary nodes, so a run writes its results straight into the tools your team already works in.
distribute
Distribute replaces file splitters.
One toggle fans a node's input into parallel jobs, split by line or by batch.
runs
Partial execution.
Hit Play on a downstream node and Trickest reuses the upstream output from your last run instead of executing it again.
scale
One node, four thousand tasks.
Distribute fans a single node across the fleet, one task per machine, with the live count on the canvas.
snapshots
Name a snapshot. Restore it later.
Every snapshot keeps its added, changed, and removed node counts so you know what you would be restoring.
debug
Stderr stands out in the console.
Error output renders in red beside stdout, so a failed run tells you what broke without digging through a log file.
schedule
Schedule from the editor.
The calendar button in the editor header turns blue once you schedule a workflow, and shows the repeat frequency.
secrets
Secrets and variables.
Variables stay readable as ${{vars.NAME}}, while secrets live in their own encrypted store as ${{secrets.NAME}} and never come back out.
data
Click Database, see the table.
One tab switch turns workflow output into a queryable findings table.
tables
Every workflow becomes a database.
The console scans recent runs, detects table-shaped output, and lists every table ready to promote.
access control
Every workspace carries its own roles.
Owners, writers and readers get exactly the access their role allows, and teams inherit that access in bulk.
Agentic updates
Workbench, agent nodes, sandboxes, and visualizations
workbench
One IDE per workflow.
Each workflow gets its own tab strip: editor, tables, query, files, terminal, and changes, all scoped to that one workflow.
plans and subagents
The agent delegates a step.
Plans persist to a file across turns, and any step can hand off to subagents that run in parallel and report back.
tasks
The agent runs on a schedule.
Save a prompt as a task, pick a cadence and a model, and read the run history it leaves behind.
ai builder
The agent builds the workflow.
Describe the scan and watch the agent place each step on your canvas and wire it up.
agent nodes
Agent nodes.
Put reasoning on the graph next to deterministic steps, so a run can triage its own results and still finish as one pipeline.
sandbox
The agent runs on a real machine.
Every sandbox is its own instance, not a microVM slice, so the agent edits files, runs commands, and builds a live preview.
visualizations
Charts in the chat.
The agent drops a chart, table, or JSON card into the thread instead of walls of text.
Pricing
Community is free with your own model key. Enterprise is one annual license for fleet, databases, and apps.
Free
Community
Explore, build, and run with Trickest on your own infrastructure.
- 1 user
- 3 self-hosted execution slots
- Workflow Builder + public workflow library
- AI Agent with your own model
- CLI, SDK, API, and MCP
- 7-day data retention
Custom
Teams & Enterprise
Run and scale security operations across your team and organization.
Everything in Community, plus:
- Managed and self-hosted execution at scale
- Multiple users, teams and shared workspaces
- All Trickest Solutions with extended capabilities
- Modules and private tooling
- RBAC, SAML SSO, and audit logs
- Flexible hosting options
- Onboarding, training, and dedicated support
free for developers · enterprise custom
Pick your plan
Drive agents from the editor you already use
One MCP line into Claude Code, Cursor, or Codex. Same engine. Same graph.
trickest · integrate · Claude Code
$ curl -fsSL https://trickest.io/integrate | sh -s -- --client claude-codeSame engine outside the UI
CLI, SDK, MCP, and REST API. Same auth. Same workflows and runs.
Learn more
Architecture and limits for every surface. Open the page that answers the next question.
Get a personalized demo
See Trickest in Action
A 30-minute walkthrough. We map the platform to your stack and answer pricing and deployment questions for your environment.