Creative Agents¶
Thinklio Built-in Agent Specification Version 0.1 | March 2026
This document covers three agents in the creative cluster: - Content Agent — manages and produces content across formats and channels - Branding Agent — develops and maintains brand identity assets - Visualiser Agent — produces charts, diagrams, and visual outputs
1. Content Agent¶
1.1 Purpose¶
The Content Agent is the production agent for organisations with regular content needs — social media, blog posts, email campaigns, case studies, newsletters, and any other recurring content format. It maintains brand voice, draws on the organisation's content library and guidelines, and produces consistently on-brand content at scale.
It is related to but distinct from the Writer Agent. The Writer Agent is a general-purpose drafting agent — it writes whatever is asked of it. The Content Agent is specifically configured for an organisation's content programme: it knows the formats, knows the brand voice, knows the audience, and knows what's been published before.
1.2 UI Structure¶
Chat Tab: Content requests and briefings: - "Write a LinkedIn post about our new service launch — professional but warm, 150 words" - "Draft a case study intro for the Cares Communities project" - "What content have we published about aged care in the last month?" - "Create this week's social media schedule based on our content themes"
Calendar Tab: Content calendar — planned and published content across channels, with status (planned, drafted, in review, published). Content created by the agent appears here. Integrates with publishing platforms where connected.
Library Tab: Archive of all produced content — searchable, tagged by format, channel, theme, and date. Useful for repurposing existing content and avoiding repetition.
Brand Tab: The brand guidelines and voice profile that govern all content produced by this agent. Admins manage this; users can view it to understand the standards applied.
1.3 Agent Capabilities¶
| Capability | Description |
|---|---|
| Produce content | Draft content in any configured format and channel |
| Repurpose | Transform existing content for a different format or channel (e.g. blog → LinkedIn post, report → summary email) |
| Adapt voice | Apply org voice profile and any channel-specific tone variations |
| Content calendar | Plan and schedule content across channels for a period |
| Research-backed content | Delegate to Research Agent for source material before writing |
| Consistency check | Review a piece of content against brand guidelines and flag deviations |
| Tag and categorise | Classify content by theme, format, and audience for library organisation |
| Performance note | When integrated with publishing platform analytics, surface what content has performed well and incorporate patterns |
1.4 Format Templates¶
The Content Agent maintains format templates for each content type the organisation uses:
ContentTemplate
├── name string (e.g. "LinkedIn post", "Case study", "Email campaign")
├── channel string
├── word_range [min, max]
├── structure string[] (sections or structural elements required)
├── tone_notes string (channel-specific tone guidance)
├── prohibited string[] (what not to include in this format)
└── examples string[] (reference examples from the content library)
1.5 Brand Voice Integration¶
The Content Agent draws on the org voice profile configured in the workspace (the same profile used by the Writer Agent). For content-specific contexts, it also reads: - Channel tone notes (LinkedIn is more professional than Instagram; email is more personal than a blog) - Audience variations (customer-facing vs. industry peers vs. internal) - Format constraints (social media character limits, email subject line best practice)
When producing content, it applies org voice as the foundation and channel/audience/format constraints as modifiers.
1.6 Content Calendar¶
The Content Agent can produce and maintain a content calendar:
- Given themes and publishing frequency, it suggests a content plan for a period
- Each calendar item has: topic, format, channel, target publish date, assigned keywords, and draft status
- Calendar items become tasks in Taskmaster for the content team
- When a draft is completed, the calendar item updates to "in review"
- Publishing (where platform integration exists) updates the item to "published"
1.7 Inter-Agent Behaviour¶
- Writer Agent handles complex long-form drafting when the Content Agent's brief exceeds its own scope
- Research Agent is invoked when content requires source material
- Fact Checker Agent reviews content claims before publication
- Report Writer Agent produces PDFs for downloadable content assets
- Branding Agent is the source of brand guidelines and visual direction
1.8 Use Cases¶
UC-1: Social media batch production A marketing manager asks the Content Agent to produce two weeks of LinkedIn posts based on the content themes document. The agent reviews the themes, produces 10 draft posts applying the brand voice and LinkedIn format template, and populates them into the content calendar. The manager reviews, edits two, and approves the rest.
UC-2: Report repurposing A research report has been produced by the Briefing Agent. The Content Agent is asked to repurpose it into: an executive summary email (300 words, email template), a LinkedIn article (800 words, thought leadership tone), and three social media posts (short, link-forward). It produces all three, each adapted for its channel.
UC-3: Brand consistency audit The Content Agent reviews 10 recently published pieces against the brand voice profile. It flags three for inconsistent tone and one for a prohibited phrase. It produces a brief summary with specific citations and suggested rewrites.
2. Branding Agent¶
2.1 Purpose¶
The Branding Agent helps develop, document, and maintain brand identity. It works through an interactive process to define or refine brand elements — name, positioning, visual identity direction, voice, and values — and produces a brand specification document that the rest of the platform (particularly the Content Agent) can use as a reference.
It is creative in a directed way: it generates options, asks questions, and iterates rather than producing a single output. The result is not a finished logo (image generation is separate) but a complete, actionable brief that could be handed to a designer or used to configure content agents consistently.
2.2 UI Structure¶
Chat Tab: The primary interface for brand development — interactive, iterative, question-and-answer.
Brand Studio Tab: The live brand specification document, built up through the chat interaction. Sections fill in as the conversation progresses. Each section can be edited directly or refined through chat.
Assets Tab: Brand output documents and any uploaded reference assets (existing logos, colour samples, etc.).
2.3 Brand Development Process¶
The Branding Agent works through a structured process, adapting to what the user already has:
Stage 1: Discovery
→ What is this for? (product, company, event, personal brand)
→ What already exists? (existing assets, constraints)
→ Who is the audience?
→ What should it feel like? What should it definitely NOT feel like?
Stage 2: Positioning
→ Core purpose and values
→ Competitive differentiation
→ Brand personality (5 adjectives)
Stage 3: Naming (if needed)
→ Name options with rationale
→ Domain and trademark consideration notes
Stage 4: Visual direction
→ Colour palette (primary and secondary, with hex values and rationale)
→ Typography direction (primary and secondary typefaces with usage notes)
→ Visual style descriptors
→ Logo direction (conceptual description, not image)
Stage 5: Voice and tone
→ Writing style guide (tone, register, preferred constructions, prohibited language)
→ Example passages for different contexts
Stage 6: Brand specification document
→ Complete structured document consolidating all decisions
→ Suitable for designer handoff or internal configuration
The agent does not need to complete all stages in one session. Users can return to continue, and the Brand Studio tab shows progress.
2.4 Agent Capabilities¶
| Capability | Description |
|---|---|
| Discovery interview | Ask structured questions to understand the brand context |
| Generate name options | Produce naming candidates with rationale |
| Colour palette generation | Develop a colour palette with semantic reasoning |
| Typography direction | Recommend typeface pairings with usage notes |
| Voice profile development | Produce a detailed writing style guide |
| Brand document production | Compile all elements into a structured specification |
| Competitive research | Delegate to Research Agent for competitive landscape context |
| Consistency review | Review existing content or assets against the developed brand spec |
2.5 Output: Brand Specification Document¶
The final output of a branding engagement is a structured document:
BrandSpec
├── brand_name string
├── tagline string | null
├── purpose string (why this brand exists)
├── audience string (who it is for)
├── personality string[] (5 adjectives)
├── positioning string (one paragraph)
├── colours
│ ├── primary[] Colour (hex, name, usage)
│ └── secondary[] Colour
├── typography
│ ├── primary TypefaceDef (name, weights, use cases)
│ └── secondary TypefaceDef
├── voice_profile VoiceProfile (same structure as org voice profile)
├── logo_direction string (conceptual description)
├── example_passages string[] (3–5 examples showing the brand voice in use)
└── constraints string[] (what the brand must never do or be)
This document is stored in the media system and can be imported as the workspace's org voice profile, making the brand spec immediately usable by the Content Agent and Writer Agent.
2.6 Use Cases¶
UC-1: New product brand A founder is launching a new app. They have a rough idea but no brand. They open the Branding Agent and answer the discovery questions. Over 45 minutes of back-and-forth, they develop: a shortlisted name, a colour palette, typography direction, a voice profile, and a logo concept description. The agent produces the brand spec document. The founder downloads it as a PDF to share with their designer.
UC-2: Brand refresh An existing brand needs updating. The user uploads the current brand guidelines and asks the Branding Agent to review and suggest a refresh. The agent analyses the existing brand, identifies areas that feel dated or inconsistent, and proposes specific updates — new secondary colour, updated typography recommendation, revised tone guidance — with rationale for each change.
3. Visualiser Agent¶
3.1 Purpose¶
The Visualiser Agent produces visual outputs from data, descriptions, and structured content. Charts, workflow diagrams, org charts, timelines, relationship maps — anything that communicates more effectively as a picture than as prose.
It is a specialist delegate, typically invoked by the Report Writer, Briefing Agent, or Data Agent when a visual would strengthen an output. It can also be used standalone by users who have data and want a chart.
3.2 Design Decisions (resolved)¶
These were flagged as open in the starter catalogue. The resolutions:
Output format by visual type: - Workflow and process diagrams: Mermaid markup (platform renders via Mermaid.js) - Data charts: Vega-Lite declarative spec (platform renders via Vega) - Custom infographics and relationship maps: SVG (agent generates directly) - Org charts: Mermaid (using graph syntax) - Timelines: Vega-Lite (using a timeline spec pattern)
Rendering responsibility: The platform provides client-side rendering for Mermaid and Vega-Lite. SVG is directly renderable. The agent never produces image files — it produces markup and specs that the platform renders. This keeps outputs editable and re-renderable without re-invoking the agent.
Interactivity (v1): Static only. The rendered output is not interactive. Vega-Lite specs support interactivity by nature, but v1 treats them as static renders. Interactive charts are v2.
3.3 UI Structure¶
Chat Tab: Visual requests: - "Create a bar chart of the quarterly revenue figures from this table" - "Draw a workflow diagram for the support triage process" - "Generate a relationship map showing the main stakeholders in this project" - "Make a timeline of the key regulatory changes from 2020 to 2026"
Gallery Tab: All produced visualisations, browsable and re-usable. Each item shows the visual, its type, the source data or description, and the output format. Visualisations can be re-rendered with updated data or style parameters.
3.4 Agent Capabilities¶
| Capability | Description |
|---|---|
| Data chart | Bar, line, area, scatter, pie — from tabular data |
| Workflow diagram | Process flows from natural language description or structured steps |
| Org chart | Hierarchy from structured data or description |
| Timeline | Chronological events from structured data |
| Relationship map | Entity-relationship diagrams |
| Infographic | Custom structured visual from a content brief |
| Apply brand style | Use org brand colours and typography direction from the Brand spec |
| Export spec | Return the raw Mermaid/Vega-Lite/SVG for external use |
3.5 Style Integration¶
When a brand spec exists in the workspace, the Visualiser Agent applies brand colours to charts and diagrams by default. This ensures that visual outputs from reports and briefings are consistently on-brand without requiring manual styling.
3.6 Inter-Agent Behaviour¶
- Report Writer invokes the Visualiser Agent to produce figures for reports, replacing image placeholder tags
- Briefing Agent uses it for visual summaries and data charts in briefing documents
- Data Agent invokes it to visualise structured data outputs
- Content Agent uses it for infographics in content pieces
3.7 Use Cases¶
UC-1: Revenue chart for report
The Report Writer is assembling a quarterly business review PDF. It encounters an image placeholder: {{IMAGE: suggestion="Bar chart of Q1–Q4 revenue by product line" placement="after_section_2"}}. It delegates to the Visualiser Agent with the data from the report and the placeholder description. The Visualiser Agent produces a Vega-Lite spec for a grouped bar chart using the org brand colours. The Report Writer embeds the rendered output.
UC-2: Workflow diagram from description A process designer asks the Visualiser Agent to "draw the support triage workflow." They describe the steps in plain English. The agent translates this into a Mermaid flowchart, renders it, and presents it. The user asks for two changes — the agent updates the spec and re-renders.
UC-3: Standalone data visualisation A data analyst pastes a CSV of sales figures and asks for a line chart showing trend over 12 months with a 3-month moving average. The agent produces a Vega-Lite spec implementing the chart with the moving average calculation inline, renders it, and returns the spec for the analyst to embed in their own tools.
4. Open Questions Across Creative Agents¶
- The Content Agent's calendar integration depends on publishing platform connections (social media schedulers, CMS). These are external integrations that need to be registered via the Integration API. The agent should function without them (producing content for manual publishing) but adds significant value when connected.
- The Branding Agent produces a Brand Specification document that can be imported as an org voice profile. The mechanism for this import — and the format compatibility between the BrandSpec output and the VoiceProfile input expected by the Writer Agent — needs to be defined.
- The Visualiser Agent's Vega-Lite and Mermaid rendering requires client-side JavaScript libraries. For PDF output via DocRaptor, these need to be rendered server-side first. The render pipeline for visual elements in PDF reports is a non-trivial engineering concern.
Previous: Business Intelligence Agents | Next: Utility Agents (Chat, Agent Studio)