Krea has opened beta access to Krea Bots, a set of named AI agents that each take on a defined creative task, carry approved preferences over from one project to the next, and hand work between one another inside a shared chat. Teams accepted into the beta will receive $1,000 in credits to try the full feature set.
The product is organized like a team chat. A user assigns a task by mentioning a specialist with @, and the agents then coordinate the steps that follow. Krea describes each bot as an AI teammate with its own role, identity, memory, tools, and isolated compute, and says it used the system to produce its recent marketing output.
A nine-agent production line
Every bot holds a defined position in the workflow. North acts as creative director: it reads the brief, plans the work, and routes tasks to the right specialists. A bot that is mentioned completes its assignment, then passes the next step to the relevant role.
- North: plans projects and briefs the other bots
- Lumen: generates images and product shots
- Doodle: creates storyboard sketches
- Voxel: builds 3D models
- Glide: produces motion graphics and animated loops
- Snip: edits videos and creates cutdowns
- Bloop: designs and mixes sound
- Ping: writes captions, schedules posts, and manages social channels
- Serif: checks brand compliance and logo use
On its landing page, Krea walks through the workflow using a product launch. North proposes three directions, Lumen generates 20 stills on travertine, Voxel models the product, Glide builds a seamless rotation, Bloop mixes room tone, Snip cuts a 15-second spot, and Ping publishes it across four channels. Krea puts the example at 43 hours of bot work against seven minutes of human input, although it does not disclose a benchmarking methodology.
What runs inside each bot
Each agent combines a reasoning model, a persistent brief, sandboxed compute, a visual review loop, and connected tools. The sandbox gives the agent an isolated environment in which to work, while the visual loop inspects images and references before the bot responds.
Krea lists Slack for requests and updates, Meta Ads for campaign launches, Figma for layouts, a browser for research, and its own image, editing, motion, sound, and 3D generators. Tool access lets a bot go beyond producing a single asset and take on parts of the surrounding production workflow.
Memory with clear boundaries
A three-layer memory model sets what each bot knows before it starts a task:
| Memory layer | What it stores | Where it applies |
|---|---|---|
| Craft | Approved lessons from feedback, such as keeping breakfast captions short | The individual specialist |
| Project | Pinned instructions and context, such as shooting at golden hour | Every bot in the current chat |
| Workspace | House rules, including logo placement and brand conventions | All bots and projects in the workspace |
Lessons that teammates propose need user approval before they enter persistent memory. That approval step gives teams a way to keep temporary feedback or mistaken instructions from turning into standing rules.
How Krea assembled the stack
Bots consolidates capabilities that Krea introduced across several earlier releases:
- August, Slack integration: Krea brought its creative tools into Slack through an adaptive agent that learns team preferences.
- September, Krea Agents: the company introduced orchestration across more than 60 models, including Flux, Veo 3.1, Kling, Runway, Luma, and Krea 2. Large language models from providers including Anthropic handle planning, while Krea and third-party systems perform generation.
- October, Krea 3: Krea released its image editor at $0.03 for a 1K image and $0.06 for a 2K image.
The shared chat packages those model-routing and generation capabilities into named roles with separate responsibilities and memories. Teams can assign ownership of each production step while keeping project-wide instructions and workspace rules in place.
The market for AI creative crews
Adobe's Firefly agents, Canva's Magic Studio, and Runway's workflow tools are expanding across the same campaign pipeline, from asset generation through editing and distribution. These products aim to coordinate multiple models and tasks inside a single creative environment.
Krea differentiates Bots through its team-chat interface and scoped memory. The structure mirrors common agency and in-house workflows, where creative direction, image production, motion, sound, editing, and distribution belong to distinct roles.
Beta access and open questions
Teams can apply through the beta waitlist. According to Krea, accepted applicants receive $1,000 in credits and access to the full feature set.
Bots-specific pricing and a general availability date have not been published. The practical value of the credit grant will depend on how Krea meters model calls, compute, storage, and connected-tool actions.
Production evaluations should examine several operational details:
- permission boundaries for Slack, Figma, Meta Ads, and browser access
- controls for viewing, editing, approving, and deleting stored memories
- handoff behavior when a bot fails or produces an unusable result
- usage logs, approval records, and cost attribution by bot or project
- API access, asset export, and integration with existing production systems
- data retention, model-provider routing, and workspace isolation
For existing Krea users, the beta adds a structured layer on top of the company's model orchestration and creative tools. For new teams, the credits leave room to test whether role-based agents can complete a campaign while preserving brand rules, human approvals, and predictable costs.