Skip to content

replit.md

Latest commit

 

History

History
70 lines (52 loc) · 4.39 KB

replit.md

File metadata and controls

70 lines (52 loc) · 4.39 KB

Recruiting Intelligence Console

An open-source, AI-native recruiting intelligence framework — a multi-agent recruiting cognition system with persistent memory, probabilistic reasoning, graph intelligence, specialized agents, declarative workflows, reusable skills, and built-in explainability and auditability. This is not an ATS, CRM, CV matcher, summarizer, or chatbot.

Run & Operate

  • pnpm --filter @workspace/api-server run dev — API server (runtime + introspection endpoints)
  • pnpm --filter @workspace/console run dev — Recruiting Intelligence Console (operator UI)
  • pnpm run typecheck — full typecheck across all packages
  • pnpm run build — typecheck + build all packages
  • pnpm --filter @workspace/api-spec run codegen — regenerate API hooks and Zod schemas from the OpenAPI spec

Stack

  • pnpm workspaces, Node.js 24, TypeScript 5.9
  • Frontend: React + Vite + Tailwind + Shadcn UI + wouter + TanStack Query
  • API: Express 5
  • DB: PostgreSQL + Drizzle ORM (schema scaffolded; runtime uses in-memory seeds today)
  • Validation: Zod (zod/v4), drizzle-zod
  • API codegen: Orval (from OpenAPI spec)
  • LLM providers: OpenAI + Anthropic via a provider abstraction (no keys required for the framework skeleton)

Where things live

  • lib/framework/src/ — the framework core. One folder per concern, each with its own index.ts:
    • agents/, skills/, providers/, orchestrator/, workflows/, memory/, candidate_graph/, organization_graph/, reports/, registry/, shared/
  • lib/api-spec/openapi.yaml — single source of truth for the HTTP contract
  • lib/api-client-react/src/generated/ — generated React Query hooks + Zod schemas
  • artifacts/api-server/src/framework/bootstrap.ts — seeds the framework registries
  • artifacts/api-server/src/routes/ — introspection routes (system, registry, memory, graph, reports)
  • artifacts/console/src/pages/ — operator console (Overview, Agents, Skills, Providers, Workflows, Memory, Graphs, Reports)
  • ARCHITECTURE.md — philosophy, module map, and design decisions

Architecture decisions

  • Framework, not application. The system is shipped as a library (@workspace/framework) with thin runtime surfaces (the API server) and an introspection UI (the console). No pipeline stages, no kanban, no candidate workflows in the ATS sense.
  • Provider abstraction first. All model calls flow through providers/. OpenAI and Anthropic are interchangeable adapters; "custom" is supported.
  • Registries everywhere. Agents, skills, providers, and workflows are all registry-backed and introspected through /registry/* endpoints.
  • Probabilistic by default. Memory entries and reports carry a confidence ∈ [0,1]. The UI renders confidence as a first-class visual primitive.
  • Provenance is data. Memory carries sourceAgentId; workflow steps reference agentId + skillId. The UI surfaces who-produced-what.
  • Graphs are explicit. Candidate and organization knowledge are exposed as node/edge snapshots rather than relational rows.

Product

The Console is the operator face of the framework. It shows the live state of the cognition system:

  • Overview — system telemetry, recent memory, recent reports, health
  • Agents — registered agents, their role, capabilities, status, default provider
  • Skills — skills grouped by category with inputs → outputs
  • Providers — LLM provider adapters and configuration status
  • Workflows — declarative workflows and their ordered (agent · skill) steps
  • Memory — persistent memory entries with scope, key, summary, confidence, provenance
  • Candidate / Organization Graph — node-link snapshots with edge relations + adjacency fallback
  • Reports — synthesised intelligence briefs with kind and confidence

User preferences

Populate as you build — explicit user instructions worth remembering across sessions.

Gotchas

  • Runtime currently uses in-memory seeds in bootstrap.ts. No DATABASE_URL is required for the skeleton to run.
  • Do not run pnpm dev at the workspace root. Use workflows or pnpm --filter <pkg> run dev.
  • Regenerate hooks/schemas after editing lib/api-spec/openapi.yaml: pnpm --filter @workspace/api-spec run codegen.

Pointers

  • See the pnpm-workspace skill for workspace structure, TypeScript setup, and package details
  • See ARCHITECTURE.md for the framework's philosophy and module-by-module map