# AskMyDocs

> Enterprise RAG + canonical knowledge graph + agentic tool use — self-hostable, MIT licensed.

## Docs

- [The admin cockpit](https://doc.askmydocs.padosoft.com/admin-panel.md): The React admin SPA at /app/admin — KPI dashboard, users + roles + RBAC, the canonical KB explorer with inline editor and graph viewer, the log viewer, the whitelisted maintenance runner, AI insights, and every cross-mounted sister-admin.
- [AI providers](https://doc.askmydocs.padosoft.com/ai-providers.md): Five interchangeable providers for chat and embeddings — OpenAI, Anthropic, Gemini, OpenRouter (raw HTTP) and Regolo (SDK adapter) — configured separately, with a dimension-safety auto-select.
- [The anti-hallucination firewall](https://doc.askmydocs.padosoft.com/anti-hallucination-firewall.md): The non-negotiable trust ranking — human-accepted > auto > raw — that lets AskMyDocs ship a self-compiling knowledge tier without ever letting machine output masquerade as human-vouched truth.
- [Auto-Wiki engine](https://doc.askmydocs.padosoft.com/architecture/auto-wiki-engine.md): The self-compiling knowledge tier — how the LLM enriches frontmatter, infers graph edges, synthesises concept pages and navigates multi-hop, all quarantined behind the reranker firewall as a second-class generation_source='auto' tier.
- [Canonical knowledge graph](https://doc.askmydocs.padosoft.com/architecture/canonical-graph.md): The typed knowledge graph — 9 node types, 10 weighted edge types, project-scoped composite foreign keys, provenance tracking, and the CanonicalIndexerJob that projects it from markdown frontmatter and wikilinks.
- [Database schema](https://doc.askmydocs.padosoft.com/architecture/database-schema.md): The PostgreSQL schema — knowledge documents and chunks, the embedding cache, chat history, the canonical graph tables, and the immutable audit trail, with their columns, indexes, and uniqueness constraints.
- [Architecture decisions](https://doc.askmydocs.padosoft.com/architecture/decisions.md): The curated narrative of the load-bearing architectural decisions — why the canonical layer, the graph, the human-gated promotion boundary, sister-package extraction, and the auto-wiki tier exist, with links to the full ADR records.
- [Ingestion pipeline](https://doc.askmydocs.padosoft.com/architecture/ingestion-pipeline.md): The deep design of ingestion — two entry points converging on one execution path, SHA-256 idempotency, section-aware fence-aware chunking, the embedding cache with batch dedup, and the canonical branch.
- [Architecture overview](https://doc.askmydocs.padosoft.com/architecture/overview.md): The end-to-end system design: the chat request lifecycle, the ingestion fan-in, the component map, and the load-bearing architectural decisions and their rationale.
- [Retrieval pipeline](https://doc.askmydocs.padosoft.com/architecture/retrieval-pipeline.md): The deep design of hybrid retrieval — over-retrieval, the reranker's weighted-sum fusion, the canonical boost/penalty trust gradient, graph expansion, and rejected-approach injection — culminating in the typed SearchResult.
- [Security & threat model](https://doc.askmydocs.padosoft.com/architecture/security-and-threat-model.md): The security architecture — multi-tenant isolation via explicit forTenant scoping, RBAC with five roles, atomic single-use confirm tokens for destructive commands, immutable audit trails, and the feature-flag discipline that keeps both states safe.
- [Auto-Wiki — the self-compiling tier](https://doc.askmydocs.padosoft.com/auto-wiki.md): A second-class 'auto' knowledge tier the system builds and maintains itself — enrichment, graph materialisation, concept synthesis, indexing, linting, agentic navigation, and cross-model review — all behind the human > auto > raw firewall.
- [Canonical knowledge & promotion](https://doc.askmydocs.padosoft.com/canonical-and-promotion.md): The typed canonical layer, the nine artifact types, and the human-gated three-stage promotion pipeline (ADR 0003) that keeps the LLM at 'draft'.
- [Chat & retrieval](https://doc.askmydocs.padosoft.com/chat-and-retrieval.md): A chat turn end to end — hybrid retrieval, reranker fusion, graph expansion, rejected-approach injection, the typed prompt, grounded citations, and the refusal contract.
- [Configuration](https://doc.askmydocs.padosoft.com/configuration.md): How AskMyDocs is configured — the env → config/*.php → service layering, and the load-bearing knobs for retrieval, reranking, the canonical layer, the graph, the refusal gate, and multi-tenancy.
- [Universal connectors](https://doc.askmydocs.padosoft.com/connectors.md): Plug AskMyDocs into Google Drive, Notion, OneDrive, Evernote, Microsoft Fabric, Confluence and Jira — OAuth in, source-aware chunked + cited out — via a standalone connector-package framework.
- [Core concepts](https://doc.askmydocs.padosoft.com/core-concepts.md): The mental model behind AskMyDocs: the canonical layer, the typed knowledge graph, evidence tiers, hybrid retrieval, the anti-hallucination firewall, and multi-tenancy.
- [Evidence & Risk Review](https://doc.askmydocs.padosoft.com/evidence-risk-review.md): An answer-grounding risk firewall — a budget-bounded sweep that labels source evidence tiers and scores per-claim risk verdicts (keep / soften / flag / remove) into a tenant-scoped review log. Integrated tri-surface from padosoft/laravel-evidence-risk-review.
- [Grounding & evidence tiers](https://doc.askmydocs.padosoft.com/grounding-and-evidence-tiers.md): How AskMyDocs keeps answers grounded in retrieved, cited context — and how the evidence-tier axis records what KIND of source a claim rests on.
- [Ingesting documents](https://doc.askmydocs.padosoft.com/ingestion.md): How content enters AskMyDocs — the two entry points, the single idempotent execution path, chunking, embeddings, and the canonical branch.
- [Installation](https://doc.askmydocs.padosoft.com/installation.md): Prerequisites, database setup with pgvector, environment configuration, and the embedding-dimension contract.
- [Institutional memory & anti-repetition](https://doc.askmydocs.padosoft.com/institutional-memory.md): Why AskMyDocs treats the KB as institutional memory — a retrieval-time knowledge graph plus a firewall that stops the LLM re-proposing approaches your team already rejected.
- [Introduction](https://doc.askmydocs.padosoft.com/introduction.md): AskMyDocs — enterprise RAG + a typed canonical knowledge graph + agentic tool use, self-hostable and MIT-licensed.
- [KITT — the embeddable agent](https://doc.askmydocs.padosoft.com/kitt-widget.md): KITT (Knowledge Interface Tour Toolkit) is a one-script-tag embeddable, page-aware, agentic assistant for any website: grounded answers with citations, reads the page, and (when allowed) drives it.
- [MCP server](https://doc.askmydocs.padosoft.com/mcp-server.md): Expose the canonical knowledge base as Model Context Protocol tools — the enterprise-kb server (25 tools) consumable by Claude Desktop, Claude Code, Cursor and any MCP client, plus token-gated access.
- [Multi-tenant isolation](https://doc.askmydocs.padosoft.com/multi-tenant-isolation.md): How AskMyDocs guarantees that one customer can never read another's data — TenantContext, BelongsToTenant (write-side auto-stamp + explicit forTenant() reads), project-scoped composite FKs, and the architecture tests that gate it.
- [PII redaction & compliance](https://doc.askmydocs.padosoft.com/pii-and-compliance.md): Field-level PII redaction at every persistence boundary (GDPR-grade, default-off, granular per touch-point) plus the EU AI Act compliance pack — DSAR, bias monitoring, risk register, consent + disclosure, audit evidence.
- [Quickstart](https://doc.askmydocs.padosoft.com/quickstart.md): From a fresh clone to your first grounded, cited answer in about five minutes.
- [ADR index](https://doc.askmydocs.padosoft.com/reference/adr-index.md): The numbered index of Architecture Decision Records — the durable record of every load-bearing decision, from the canonical layer to the Auto-Wiki tier.
- [HTTP API reference](https://doc.askmydocs.padosoft.com/reference/api.md): The REST/JSON API surface — authentication, KB chat & search, ingestion, the human-gated promotion pipeline, conversations, the embeddable widget, and the RBAC-gated admin endpoints.
- [CLI reference](https://doc.askmydocs.padosoft.com/reference/cli.md): Every Artisan command — ingestion, deletion, promotion, the canonical graph, the Auto-Wiki engine, retention sweeps, insights, evaluation, and the MCP/widget helpers — with exact signatures.
- [Scheduler & maintenance](https://doc.askmydocs.padosoft.com/scheduler-and-maintenance.md): The config-driven task scheduler — every nightly job, its default cron, and its per-slot enable/cron env overrides — plus the maintenance commands for retention, graph rebuild, and orphan cleanup.
- [Self-hosting](https://doc.askmydocs.padosoft.com/self-hosting.md): Run AskMyDocs on your own infrastructure — requirements, a step-by-step install, the PostgreSQL + pgvector setup, the SPA build, and the worker + scheduler processes that keep it alive.
- [Sister packages](https://doc.askmydocs.padosoft.com/sister-packages.md): The standalone padosoft/* Composer packages AskMyDocs builds on — each MIT, standalone-agnostic, and independently reusable: AI providers, PII redaction, saga engine, eval harness, connectors, MCP pack, and AI Act compliance.
- [Troubleshooting](https://doc.askmydocs.padosoft.com/troubleshooting.md): Diagnose the common failure modes — the embedding-dimension gotcha, the health-check surface, silent delete no-ops, spurious refusals, an empty graph, and a stalled ingestion queue.