th0th

Ancient knowledge keeper for modern code

Semantic search with 98% token reduction for AI assistants.

Como reduzi 98% do uso de contexto (e custos) de IA no meu workflow / How I reduced AI context usage (and costs) by 98% in my workflow https://www.tabnews.com.br/S1LV4/como-reduzi-em-98-por-cento-o-uso-de-contexto-e-os-custos-de-ia-no-meu-workflow

---

Quick Start

One-line install (recommended)

curl -fsSL https://raw.githubusercontent.com/S1LV4/th0th/main/install.sh | bash

Installs interactively. Three modes:

Mode	Requires	Best for
Docker (default)	Docker	Production, quick start
Docker build	Docker + Git	Custom builds, local changes
Source	Git + Bun	Development, contributors

Non-interactive (CI/scripted):

# Docker mode, custom port, skip start
TH0TH_MODE=docker TH0TH_API_PORT=4000 TH0TH_NO_START=1 \
  curl -fsSL https://raw.githubusercontent.com/S1LV4/th0th/main/install.sh | bash

Manual setup (from source)

# 1. Clone and install
git clone https://github.com/S1LV4/th0th.git
cd th0th
bun install

# 2. Setup (100% offline with Ollama)
./scripts/setup-local-first.sh
# - Installs/starts Ollama
# - Pulls bge-m3 embedding model (1024 dimensions)
# - Creates .env with defaults
# - Runs bun run diagnose to validate the stack

# 3. Build and start
bun run build
bun run start:api

Verify: curl http://localhost:3333/health

Tip: Run bun run diagnose at any time to validate Ollama connectivity, database access, embedding generation, and migration status.

---

Integration

OpenCode (recommended)

File: ~/.config/opencode/opencode.json

Via MCP package:

{
  "mcp": {
    "th0th": {
      "type": "local",
      "command": [
        "bunx",
        "@th0th-ai/mcp-client"
      ],
      "environment": {
        "TH0TH_API_URL": "http://localhost:3333"
      },
      "enabled": true
    }
  }
}

Via Plugin:

{
  "plugin": ["@th0th-ai/opencode-plugin"]
}

From source (development):

{
  "mcpServers": {
    "th0th": {
      "type": "local",
      "command": ["bun", "run", "/path/to/th0th/apps/mcp-client/src/index.ts"],
      "enabled": true
    }
  }
}

VSCode / Antigravity

Create .vscode/mcp.json in your workspace:

{
  "servers": {
    "th0th": {
      "command": "bunx",
        "args": ["@th0th-ai/mcp-client"],
      "env": {
        "TH0TH_API_URL": "http://localhost:3333"
      }
    }
  }
}

Or run ./scripts/setup-vscode.sh for automatic configuration.

Docker

{
  "mcpServers": {
    "th0th": {
      "type": "local",
      "command": ["docker", "compose", "run", "--rm", "-i", "mcp"],
      "enabled": true
    }
  }
}

---

Available Tools

Indexing & Search

Tool	Description
th0th_index	Index a project directory with semantic embeddings
th0th_index_status	Poll background indexing job progress
th0th_search	Hybrid semantic + keyword search with RRF ranking. Supports responseMode=enriched for full content + imports + parentSymbol in one call
th0th_reindex	Force full reindex after a large refactor
th0th_reset_project	Delete all indexed data for a project (vectors, symbols, memories)
th0th_list_projects	List all indexed projects with status and file counts
th0th_project_map	One-shot project summary: stats, top files by PageRank, symbol distribution

Symbol Graph

Tool	Description
th0th_search_definitions	Find function/class/type definitions by name
th0th_get_references	Find all usages of a symbol across the project
th0th_go_to_definition	Jump to definition with file + line context
th0th_symbol_snippet	Get raw code snippet by file + line range
th0th_read_file	Read a file with symbol metadata and imports

Memory

Tool	Description
th0th_remember	Store important information in persistent memory
th0th_recall	Semantic search over stored memories
th0th_memory_list	Browse memories by type/importance (audit mode)
th0th_compress	Compress context (keeps structure, removes detail)
th0th_optimized_context	Search + compress in one call (max token efficiency)
th0th_analytics	Usage patterns, cache performance, metrics

Synapse (Cognitive Layer)

Synapse is an optional post-retrieval modulation layer that improves result quality over a session by tracking task context, agent affinity, and working-memory. Enable by creating a session and passing sessionId to th0th_search.

Tool	Description
th0th_synapse_session	Create/resume a cognitive session scoped to a task
th0th_synapse_prime	Seed working-memory buffer with recalled memories
th0th_synapse_access	Record file access to boost that file in future searches

---

Search Quality Tuning

Environment variables for fine-tuning retrieval (all optional):

Variable	Default	Description
SEARCH_DISABLE_KEYWORD	false	Pure vector-only mode (+44% MRR on NL→code)
RRF_KEYWORD_BOOST	2.5	Keyword weight multiplier for code queries
RRF_VECTOR_WEIGHT	0.3	Vector similarity weight in final score blend
RRF_MAX_CHUNKS_PER_FILE	2	Diversity cap — prevents one file monopolising results
SEARCH_MIN_SCORE	0.3	Score threshold below which results are dropped
OLLAMA_EMBED_DELAY_MS	0	Delay between Ollama embed calls (set >0 for CPU)

---

REST API

# Development
bun run dev:api

# Production
bun run start:api

Swagger docs: http://localhost:3333/swagger

Endpoints

# Index a project
curl -X POST http://localhost:3333/api/v1/project/index \
  -H "Content-Type: application/json" \
  -d '{"projectPath": "/home/user/my-project", "projectId": "my-project"}'

# Search
curl -X POST http://localhost:3333/api/v1/search/project \
  -H "Content-Type: application/json" \
  -d '{"query": "authentication", "projectId": "my-project"}'

# Store memory
curl -X POST http://localhost:3333/api/v1/memory/store \
  -H "Content-Type: application/json" \
  -d '{"content": "Important decision...", "type": "decision"}'

# Compress context
curl -X POST http://localhost:3333/api/v1/context/compress \
  -H "Content-Type: application/json" \
  -d '{"content": "...", "strategy": "code_structure"}'

---

Configuration

Config file: ~/.config/th0th/config.json (auto-created on first run)

Quick Config Commands

# Show current configuration
npx @th0th-ai/mcp-client --config-show

# Show config file path
npx @th0th-ai/mcp-client --config-path

# Show config directory
npx @th0th-ai/mcp-client --config-dir

# Initialize configuration
npx @th0th-ai/mcp-client --config-init

# Show help
npx @th0th-ai/mcp-client --help

Embedding Providers

Provider	Model	Cost	Quality
Ollama (default)	qwen3-embedding, bge-m3, nomic-embed-text	Free	Good-Excellent
Mistral	mistral-embed, codestral-embed	$$	Great
OpenAI	text-embedding-3-small	$$	Great

Advanced Configuration

For detailed configuration management, use the config CLI:

# Initialize with specific provider
npx @th0th-ai/mcp-client --config-init                          # Ollama (default)
npx @th0th-ai/mcp-client --config-init --mistral your-api-key   # Mistral
npx @th0th-ai/mcp-client --config-init --openai your-api-key    # OpenAI

# Switch provider
npx @th0th-ai/mcp-client --config-init --mistral your-api-key
npx @th0th-ai/mcp-client --config-init --ollama-model qwen3-embedding

# Set specific configuration values
npx @th0th-ai/mcp-client --config-set embedding.dimensions 4096

---

Scripts

Command	Description
bun run build	Build all packages
bun run dev	Development (all apps)
bun run dev:api	REST API with hot reload
bun run dev:mcp	MCP server with watch
bun run start:api	Start REST API
bun run start:mcp	Start MCP server
bun run test	Run tests
bun run lint	Lint code
bun run type-check	Type checking
bun run diagnose	Validate full stack (Ollama, database, embeddings)

---

Architecture

th0th/
├── apps/
│   ├── mcp-client/           # MCP Server (stdio)
│   ├── tools-api/            # REST API (port 3333)
│   └── opencode-plugin/      # OpenCode plugin
├── packages/
│   ├── core/                 # Business logic, search, embeddings, compression
│   └── shared/               # Shared types & utilities
└── scripts/

Component	Description
Semantic Search	Hybrid vector + keyword with RRF ranking, enriched response mode
Synapse	Post-retrieval cognitive modulation: task alignment, agent affinity, working-memory buffer
Symbol Graph	PageRank-based centrality, definitions, references, go-to-definition
Embeddings	Ollama (local) or Mistral/OpenAI API
Compression	Rule-based code structure extraction (70-98% reduction)
Memory	Persistent SQLite/PostgreSQL storage across sessions
Cache	Multi-level L1/L2 with TTL

---

License

MIT