HAAM Agent Hub
The machines at work, with their limits made visible.
This is the current registry of AI agents and copilots wired into HAAM. Each record names the system's responsibility, access, autonomy, evidence, operating surface, and the point where human authority begins.
The registry describes implemented runtime systems rather than future concepts. Access labels distinguish the live microsite, dormant full-site tools, authenticated admin workspaces, and internal APIs.
- 13
- AI systems registered
- 2
- scheduled or autonomous
- 6
- public or full-site surfaces
- 6
- private admin systems
2 systems
Autonomous operations
Scheduled or tool-using systems that can continue a bounded workflow without a live prompt. Their outputs still stop at an explicit human decision point.
Documentation Steward
Reviews bounded GitHub changes for documentation drift, prepares evidence-backed MDX updates, and opens a pull request for review.
Responsibility: Keep the public documentation aligned with the system it describes.
- Detect documentation drift
- Draft bounded documentation changes
- Open reviewable pull requests
Operating surface
Vercel Cron, GitHub comparison, and pull-request handoff
Human gate
A person reviews and merges every proposed documentation change before it can publish.
Operating proof
0 published reviews, 0 code files reviewed, and 0 documentation proposals
Code references
- app/agents/documentation-steward/page.js
- lib/docs-steward-vercel.js
Problem Radar
Searches public communities and privacy-filtered visitor questions, clusters substantially similar complaints, scores demand signals, and stores structured reports.
Responsibility: Find recurring, externally corroborated problems that may justify a product, service, or investigation.
- Search public problem sources
- Cluster recurring complaints
- Score demand and HAAM fit
Operating surface
Separate Vercel Eve project and the Problem Radar dashboard
Human gate
A person decides whether a pattern becomes a lead, project, offer, or editorial direction.
Operating proof
Daily at 05:15 UTC, with public evidence required before saving a report
Code references
- agents/problem-radar/agent/instructions.md
- agents/problem-radar/agent/schedules/daily-demand-scan.ts
4 systems
Interactive agents
User-facing systems that reason inside a live session, use approved tools when available, and return a reviewable answer, audit, brief, or workflow.
HAAM Signal Agent
Uses a stateful Pi agent runtime and specialist tools to inspect hiring, product, procurement, customer pain, GitHub, advertising, financial, technology, regulatory, and domain signals.
Responsibility: Turn scattered public market evidence into a decision-ready project brief.
- Run multi-source market scans
- Form evidence-backed hypotheses
- Prepare an approved project brief
Operating surface
Pi Agent runtime, Vercel AI Gateway, and bounded public-signal tools
Human gate
The agent can prepare a brief, but delivery of a lead requires explicit approval of that brief.
Operating proof
Sequential tool execution, source status reporting, and approval before submission
Code references
- lib/haam-signal-agent/runtime.js
- lib/haam-signal-agent/tools.js
Web Studio Assistant
Provides concise, page-aware project guidance through a rate-limited streaming chat. It helps before asking for contact details and does not take external actions.
Responsibility: Help a visitor clarify the goals, scope, budget, timeline, constraints, and content needs of a project.
- Clarify a project request
- Surface missing constraints
- Prepare a better human handoff
Operating surface
The Start page and the public streaming chat API
Human gate
The visitor chooses whether to share contact details or request a follow-up from a person.
Operating proof
Public, rate-limited, page-aware, and without forced contact capture
Code references
- app/api/chat/route.js
- components/chatbot.js
Content Audit Agent
Audits a public page or supplied draft with deterministic checks and an optional AI review, then returns scored findings and a prioritized action plan.
Responsibility: Diagnose content quality across comprehension, credibility, accessibility, findability, maintenance, and user action.
- Inspect page and draft structure
- Prioritize evidence-backed issues
- Create a revision action plan
Operating surface
Content Audit tool, guarded URL reader, and structured audit engine
Human gate
A person verifies factual accuracy, audience fit, brand judgment, and the final revision.
Operating proof
Eight scored dimensions with a deterministic fallback when AI is unavailable
Code references
- lib/content-audit-agent.js
- app/api/content-audit/route.js
Automation Architect
Converts a plain-language process into an implementation-ready blueprint with triggers, inputs, decisions, tools, ownership, risk, metrics, monitoring, and fallback behavior.
Responsibility: Turn recurring work into the smallest useful human-supervised automation loop.
- Design end-to-end workflows
- Place approval boundaries
- Define monitoring and failure behavior
Operating surface
AI Automation Studio and the blueprint API
Human gate
Consequential sending, publishing, spending, deletion, access, record mutation, and sensitive decisions require an approval step.
Operating proof
Validated workflow schema with explicit risk, fallback, metrics, and approval fields
Code references
- app/api/automation-studio/blueprint/route.js
7 systems
Internal copilots
Private research, editorial, design, sales, and discovery systems that expand coverage while keeping final judgment and consequential action with a person.
Design Direction Assistant
Selects only from approved template and theme IDs, proposes bounded design tokens, and explains the recommendation without directly changing the live site.
Responsibility: Translate an art-direction request into a restrained configuration for the existing production design system.
- Recommend a template and theme
- Propose constrained design tokens
- Explain the visual direction
Operating surface
Authenticated Design Studio and constrained template and theme catalogues
Human gate
A person reviews, previews, and applies the final design configuration.
Operating proof
Authenticated access, allow-listed IDs, normalized tokens, and bounded font scale
Code references
- app/api/admin/design/assistant/route.js
Editorial Engine
Scores proposed articles, reads bounded sources, separates evidence from inference, selects internal links, drafts long-form Markdown, and diagnoses existing performance.
Responsibility: Turn sourced signals into durable editorial opportunities, research packets, drafts, and refresh plans.
- Score editorial opportunities
- Build evidence-bound research packets
- Draft and refresh articles
Operating surface
Editorial dashboard, source collector, site index, and structured AI outputs
Human gate
A person verifies claims, chooses the argument, edits the draft, and decides whether anything publishes.
Operating proof
Source-bound research packets, explicit risks, and no automatic publishing
Code references
- lib/editorial-engine-ai.js
- app/api/admin/editorial/route.js
Opportunity Agent
Builds a professional search profile from public HAAM pages, expands adjacent role queries, discovers opportunities, scores fit, and prepares application material.
Responsibility: Find and evaluate professional opportunities that fit the available public evidence about Kris and HAAM.
- Build an evidence-based search profile
- Discover and score opportunities
- Prepare application drafts
Operating surface
Admin Radar, source adapters, scoring, storage, and application modules
Human gate
A person chooses which roles to pursue, verifies every claim, and sends each application.
Operating proof
Professional evidence only, with protected and private traits excluded
Code references
- lib/opportunity-agent-profile.js
- lib/opportunity-application-generator.js
Prospect Analyst
Combines a guarded website scan with supplied context to score website opportunity, service fit, buying trigger, budget likelihood, and contactability.
Responsibility: Evaluate whether a public organization presents a credible, timely, and reachable HAAM opportunity.
- Inspect public website signals
- Score commercial fit and timing
- Prepare an evidence-based outreach angle
Operating surface
Prospecting dashboard, public website reader, heuristic baseline, and AI enhancement
Human gate
A person verifies the trigger, budget, contact, and ethical basis before any outreach.
Operating proof
Five scored dimensions with deterministic evidence retained alongside AI analysis
Code references
- lib/prospect-analysis.js
- app/api/admin/prospects/analyze/route.js
PhD Programme Scout
Builds a doctoral search profile from public pages, maintains a source registry, ingests programme evidence, and scores opportunities across target and adjacent fields.
Responsibility: Find doctoral programmes and funded researcher roles that match the documented research direction.
- Build a doctoral search profile
- Discover and normalize programmes
- Score research and funding fit
Operating surface
PhD dashboard, programme source registry, profile builder, ingestion, and scoring
Human gate
A person verifies programme details, contacts supervisors, and submits every application.
Operating proof
Academic and professional evidence only, with no invented credentials or publications
Code references
- lib/phd-programme-profile.js
- lib/phd-programme-scoring.js
City Intelligence Enricher
Classifies local items, writes compact descriptions, explains relevance, suggests a next action, adds tags and audiences, and attaches evidence-based confidence.
Responsibility: Turn sourced local signals into consistent, useful city intelligence without inventing missing facts.
- Classify local opportunities and events
- Add practical context and actions
- Expose uncertainty through confidence
Operating surface
City intelligence API with schema validation and deterministic fallback
Human gate
Editors and readers verify dates, prices, eligibility, organizations, locations, and outcomes at the original source.
Operating proof
One result per input item, capped batches, schema validation, and attached confidence
Code references
- lib/city-intelligence-ai.js
- app/api/city-intelligence/route.js
Design Language Research Assistant
Examines public HTML, visible text, image metadata, fonts, colors, composition, and motion signals while avoiding claims about identity, personality, intent, or private life.
Responsibility: Create a cautious, evidence-led reading of the design language visible on a public portfolio page.
- Inspect observable visual signals
- Separate interface from project work
- State evidence and uncertainty
Operating surface
Graphic Designers directory and guarded public portfolio analysis
Human gate
A person treats the output as a hypothesis, checks project context, and decides how or whether to use it.
Operating proof
One-page evidence, explicit caution, confidence level, and deterministic fallback
Code references
- lib/designer-analysis.js
The HAAM standard
No invisible autonomy.
Every useful agent needs a named job, the least access it needs, inspectable evidence, visible uncertainty, reversible behavior, and a clear stop where human responsibility takes over.