Docs

Agent reference

Technical specifications for Boki's AI agents

Agent reference

Technical specifications for all Boki agents.

Global chat agent

The primary interface accessed via the orange sparkle icon.

Modes

ModeBehavior
AutoBoki decides approach based on query — may search web, analyze workspace, or combine
ResearchPrioritizes web search and external source analysis

Context

  • Has access to workspace content (plans, briefs, articles)
  • @ mentions load specific document content into context
  • Conversation history persists within session

Greeting

Initial message: "Hi! I'm Boki."

Researcher agent

Specialized agent for deep research tasks.

Capabilities

CapabilityDescription
Web searchSearches public web for relevant content
Source analysisEvaluates and summarizes sources
Competitive analysisCompares your content against external content
Gap identificationFinds topics and angles not covered by competitors

Artifacts (Beta)

ArtifactDescription
OutlinesSuggested content structure
Keyword clustersRelated terms grouped by intent
Source listsCurated sources with summaries

Limitations

  • Cannot access paywalled content
  • Web search results may not include very recent content
  • Does not write full articles

Technical review agent

Reviews content for technical accuracy.

Requirements

  • Minimum 50 words of selected text
  • Works in article editor only (inline)

Capabilities

CheckDescription
Code verificationSyntax checking, logic review
Sandbox executionCan run code to verify behavior
Fact-checkingVerifies technical claims against sources
Consistency checkEnsures terminology and examples are consistent

Output

Returns structured feedback with:

  • ✓ Verified elements
  • ⚠ Issues requiring attention
  • 💡 Suggestions for improvement

Marketer review agent

Reviews content for marketing effectiveness.

Requirements

  • Minimum 50 words of selected text
  • Works in article editor only (inline)

Evaluation criteria

CriterionDescription
StructureLogical flow, transitions, opening/closing
Funnel alignmentMatches intended stage (awareness/consideration/decision)
Source qualityClaims supported by credible sources
ClarityMain point clear, jargon explained

Output

Returns structured feedback with:

  • Identified issues
  • Specific improvement suggestions
  • No grammar or spelling checks (not the purpose)

LLM search visibility agent

Beta

Scores content against specific questions for AI visibility.

Requirements

  • Minimum 50 words of selected text
  • A specific question to score against
  • Works in article editor only (inline)

Scoring factors

FactorWeightDescription
Question alignmentHighDoes content directly answer the question?
ComprehensivenessHighThorough coverage of topic
Authority signalsMediumClear definitions, structure, citations
DifferentiationMediumUnique value not in competing content

Score interpretation

RangeMeaning
80-100Strong — likely cited as primary source
60-79Moderate — may be cited, competing content stronger
40-59Weak — unlikely cited without improvements
0-39Poor — doesn't address question adequately

Output

Returns:

  • Numeric score (0-100)
  • Strengths of the content
  • Gaps compared to competing content
  • Specific suggestions

Limitations

  • Score is predictive, not definitive
  • Different LLMs may weight factors differently
  • Does not measure traditional SEO ranking

Common limitations (all agents)

  • Agents do not write full articles
  • Agents provide advisory feedback — user decides what to act on
  • Agents cannot access your account credentials or external services
  • Agent capabilities may vary based on plan tier