cli/pkg/rag/ingest.go

217 lines
5.2 KiB
Go
Raw Normal View History

feat: wire release command, add tar.xz support, unified installers (#277) * feat(cli): wire release command and add installer scripts - Wire up `core build release` subcommand (was orphaned) - Wire up `core monitor` command (missing import in full variant) - Add installer scripts for Unix (.sh) and Windows (.bat) - setup: Interactive with variant selection - ci: Minimal for CI/CD environments - dev: Full development variant - go/php/agent: Targeted development variants - All scripts include security hardening: - Secure temp directories (mktemp -d) - Architecture validation - Version validation after GitHub API call - Proper cleanup on exit - PowerShell PATH updates on Windows (avoids setx truncation) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> * feat(build): add tar.xz support and unified installer scripts - Add tar.xz archive support using Borg's compress package - ArchiveXZ() and ArchiveWithFormat() for configurable compression - Better compression ratio than gzip for release artifacts - Consolidate 12 installer scripts into 2 unified scripts - install.sh and install.bat with BunnyCDN edge variable support - Subdomains: setup.core.help, ci.core.help, dev.core.help, etc. - MODE and VARIANT transformed at edge based on subdomain - Installers prefer tar.xz with automatic fallback to tar.gz - Fixed CodeRabbit issues: HTTP status patterns, tar error handling, verify_install params, VARIANT validation, CI PATH persistence Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> * chore: add build and release config files - .core/build.yaml - cross-platform build configuration - .core/release.yaml - release workflow configuration Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> * chore: move plans from docs/ to tasks/ Consolidate planning documents in tasks/plans/ directory. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> * fix(install): address CodeRabbit review feedback - Add curl timeout (--max-time) to prevent hanging on slow networks - Rename TMPDIR to WORK_DIR to avoid clobbering system env var - Add chmod +x to ensure binary has execute permissions - Add error propagation after subroutine calls in batch file - Remove System32 install attempt in CI mode (use consistent INSTALL_DIR) - Fix HTTP status regex for HTTP/2 compatibility Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> * feat(rag): add Go RAG implementation with Qdrant + Ollama Add RAG (Retrieval Augmented Generation) tools for storing documentation in Qdrant vector database and querying with semantic search. This replaces the Python tools/rag implementation with a native Go solution. New commands: - core rag ingest [directory] - Ingest markdown files into Qdrant - core rag query [question] - Query vector database with semantic search - core rag collections - List and manage Qdrant collections Features: - Markdown chunking by sections and paragraphs with overlap - UTF-8 safe text handling for international content - Automatic category detection from file paths - Multiple output formats: text, JSON, LLM context injection - Environment variable support for host configuration Dependencies: - github.com/qdrant/go-client (gRPC client) - github.com/ollama/ollama/api (embeddings API) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> * feat(deploy): add pure-Go Ansible executor and Coolify API integration Implement infrastructure deployment system with: - pkg/ansible: Pure Go Ansible executor - Playbook/inventory parsing (types.go, parser.go) - Full execution engine with variable templating, loops, blocks, conditionals, handlers, and fact gathering (executor.go) - SSH client with key/password auth and privilege escalation (ssh.go) - 35+ module implementations: shell, command, copy, template, file, apt, service, systemd, user, group, git, docker_compose, etc. (modules.go) - pkg/deploy/coolify: Coolify API client wrapping Python swagger client - List/get servers, projects, applications, databases, services - Generic Call() for any OpenAPI operation - pkg/deploy/python: Embedded Python runtime for swagger client integration - internal/cmd/deploy: CLI commands - core deploy servers/projects/apps/databases/services/team - core deploy call <operation> [params-json] This enables Docker-free infrastructure deployment with Ansible-compatible playbooks executed natively in Go. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> * fix(deploy): address linter warnings and build errors - Fix fmt.Sprintf format verb error in ssh.go (remove unused stat command) - Fix errcheck warnings by explicitly ignoring best-effort operations - Fix ineffassign warning in cmd_ansible.go All golangci-lint checks now pass for deploy packages. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> * style(deploy): fix gofmt formatting Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> * fix(deploy): use known_hosts for SSH host key verification Address CodeQL security alert by using the user's known_hosts file for SSH host key verification when available. Falls back to accepting any key only when known_hosts doesn't exist (common in containerized or ephemeral environments). Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> * feat(ai,security,ide): add agentic MVP, security jobs, and Core IDE desktop app Wire up AI infrastructure with unified pkg/ai package (metrics JSONL, RAG integration), move RAG under `core ai rag`, add `core ai metrics` command, and enrich task context with Qdrant documentation. Add `--target` flag to all security commands for external repo scanning, `core security jobs` for distributing findings as GitHub Issues, and consistent error logging across scan/deps/alerts/secrets commands. Add Core IDE Wails v3 desktop app with Angular 20 frontend, MCP bridge (loopback-only HTTP server), WebSocket hub, and Claude Code bridge. Production-ready with Lethean CIC branding, macOS code signing support, and security hardening (origin validation, body size limits, URL scheme checks, memory leak prevention, XSS mitigation). Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> * fix: address PR review comments from CodeRabbit, Copilot, and Gemini Fixes across 25 files addressing 46+ review comments: - pkg/ai/metrics.go: handle error from Close() on writable file handle - pkg/ansible: restore loop vars after loop, restore become settings, fix Upload with become=true and no password (use sudo -n), honour SSH timeout config, use E() helper for contextual errors, quote git refs in checkout commands - pkg/rag: validate chunk config, guard negative-to-uint64 conversion, use E() helper for errors, add context timeout to Ollama HTTP calls - pkg/deploy/python: fix exec.ExitError type assertion (was os.PathError), handle os.UserHomeDir() error - pkg/build/buildcmd: use cmd.Context() instead of context.Background() for proper Ctrl+C cancellation - install.bat: add curl timeouts, CRLF line endings, use --connect-timeout for archive downloads - install.sh: use absolute path for version check in CI mode - tools/rag: fix broken ingest.py function def, escape HTML in query.py, pin qdrant-client version, add markdown code block languages - internal/cmd/rag: add chunk size validation, env override handling Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> * fix(build): make release dry-run by default and remove darwin/amd64 target Replace --dry-run (default false) with --we-are-go-for-launch (default false) so `core build release` is safe by default. Remove darwin/amd64 from default build targets (arm64 only for macOS). Fix cmd_project.go to use command context instead of context.Background(). Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
2026-02-04 00:49:57 +00:00
package rag
import (
"context"
"fmt"
"io/fs"
"os"
"path/filepath"
"strings"
"github.com/host-uk/core/pkg/log"
)
// IngestConfig holds ingestion configuration.
type IngestConfig struct {
Directory string
Collection string
Recreate bool
Verbose bool
BatchSize int
Chunk ChunkConfig
}
// DefaultIngestConfig returns default ingestion configuration.
func DefaultIngestConfig() IngestConfig {
return IngestConfig{
Collection: "hostuk-docs",
BatchSize: 100,
Chunk: DefaultChunkConfig(),
}
}
// IngestStats holds statistics from ingestion.
type IngestStats struct {
Files int
Chunks int
Errors int
}
// IngestProgress is called during ingestion to report progress.
type IngestProgress func(file string, chunks int, total int)
// Ingest processes a directory of documents and stores them in Qdrant.
func Ingest(ctx context.Context, qdrant *QdrantClient, ollama *OllamaClient, cfg IngestConfig, progress IngestProgress) (*IngestStats, error) {
stats := &IngestStats{}
// Validate batch size to prevent infinite loop
if cfg.BatchSize <= 0 {
cfg.BatchSize = 100 // Safe default
}
// Resolve directory
absDir, err := filepath.Abs(cfg.Directory)
if err != nil {
return nil, log.E("rag.Ingest", "error resolving directory", err)
}
info, err := os.Stat(absDir)
if err != nil {
return nil, log.E("rag.Ingest", "error accessing directory", err)
}
if !info.IsDir() {
return nil, log.E("rag.Ingest", fmt.Sprintf("not a directory: %s", absDir), nil)
}
// Check/create collection
exists, err := qdrant.CollectionExists(ctx, cfg.Collection)
if err != nil {
return nil, log.E("rag.Ingest", "error checking collection", err)
}
if cfg.Recreate && exists {
if err := qdrant.DeleteCollection(ctx, cfg.Collection); err != nil {
return nil, log.E("rag.Ingest", "error deleting collection", err)
}
exists = false
}
if !exists {
vectorDim := ollama.EmbedDimension()
if err := qdrant.CreateCollection(ctx, cfg.Collection, vectorDim); err != nil {
return nil, log.E("rag.Ingest", "error creating collection", err)
}
}
// Find markdown files
var files []string
err = filepath.WalkDir(absDir, func(path string, d fs.DirEntry, err error) error {
if err != nil {
return err
}
if !d.IsDir() && ShouldProcess(path) {
files = append(files, path)
}
return nil
})
if err != nil {
return nil, log.E("rag.Ingest", "error walking directory", err)
}
if len(files) == 0 {
return nil, log.E("rag.Ingest", fmt.Sprintf("no markdown files found in %s", absDir), nil)
}
// Process files
var points []Point
for _, filePath := range files {
relPath, err := filepath.Rel(absDir, filePath)
if err != nil {
stats.Errors++
continue
}
content, err := os.ReadFile(filePath)
if err != nil {
stats.Errors++
continue
}
if len(strings.TrimSpace(string(content))) == 0 {
continue
}
// Chunk the content
category := Category(relPath)
chunks := ChunkMarkdown(string(content), cfg.Chunk)
for _, chunk := range chunks {
// Generate embedding
embedding, err := ollama.Embed(ctx, chunk.Text)
if err != nil {
stats.Errors++
if cfg.Verbose {
fmt.Printf(" Error embedding %s chunk %d: %v\n", relPath, chunk.Index, err)
}
continue
}
// Create point
points = append(points, Point{
ID: ChunkID(relPath, chunk.Index, chunk.Text),
Vector: embedding,
Payload: map[string]any{
"text": chunk.Text,
"source": relPath,
"section": chunk.Section,
"category": category,
"chunk_index": chunk.Index,
},
})
stats.Chunks++
}
stats.Files++
if progress != nil {
progress(relPath, stats.Chunks, len(files))
}
}
// Batch upsert to Qdrant
if len(points) > 0 {
for i := 0; i < len(points); i += cfg.BatchSize {
end := i + cfg.BatchSize
if end > len(points) {
end = len(points)
}
batch := points[i:end]
if err := qdrant.UpsertPoints(ctx, cfg.Collection, batch); err != nil {
return stats, log.E("rag.Ingest", fmt.Sprintf("error upserting batch %d", i/cfg.BatchSize+1), err)
}
}
}
return stats, nil
}
// IngestFile processes a single file and stores it in Qdrant.
func IngestFile(ctx context.Context, qdrant *QdrantClient, ollama *OllamaClient, collection string, filePath string, chunkCfg ChunkConfig) (int, error) {
content, err := os.ReadFile(filePath)
if err != nil {
return 0, log.E("rag.IngestFile", "error reading file", err)
}
if len(strings.TrimSpace(string(content))) == 0 {
return 0, nil
}
category := Category(filePath)
chunks := ChunkMarkdown(string(content), chunkCfg)
var points []Point
for _, chunk := range chunks {
embedding, err := ollama.Embed(ctx, chunk.Text)
if err != nil {
return 0, log.E("rag.IngestFile", fmt.Sprintf("error embedding chunk %d", chunk.Index), err)
}
points = append(points, Point{
ID: ChunkID(filePath, chunk.Index, chunk.Text),
Vector: embedding,
Payload: map[string]any{
"text": chunk.Text,
"source": filePath,
"section": chunk.Section,
"category": category,
"chunk_index": chunk.Index,
},
})
}
if err := qdrant.UpsertPoints(ctx, collection, points); err != nil {
return 0, log.E("rag.IngestFile", "error upserting points", err)
}
return len(points), nil
}