codegraph-rust/.env.example at main · Jakedismo/codegraph-rust · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
# CodeGraph env example (copy to .env). Prefer env vars over committing secrets.

# --- Project identity ---
# CODEGRAPH_PROJECT_ID=my-project
# CODEGRAPH_ORGANIZATION_ID=my-org
# CODEGRAPH_REPOSITORY_URL=https://github.com/user/repo
# CODEGRAPH_DOMAIN=example.com

# --- Embeddings ---
CODEGRAPH_EMBEDDING_PROVIDER=ollama          # ollama | openai | jina | lmstudio | onnx
CODEGRAPH_EMBEDDING_DIMENSION=1024
CODEGRAPH_EMBEDDINGS_BATCH_SIZE=64           # batch per provider call
CODEGRAPH_EMBEDDING_MODEL=qwen3-embedding:0.6B
# CODEGRAPH_CHUNK_MAX_TOKENS=1024             # override model max tokens for chunking
# CODEGRAPH_EMBEDDING_SKIP_CHUNKING=0         # set 1 to disable chunking and embed nodes directly faster decently accurate with large ctx window embedding modelse

# --- Reranking (optional) ---
# CODEGRAPH_RERANKING_PROVIDER=jina           # jina
# CODEGRAPH_RERANKING_MODEL=jina-reranker-v3
# CODEGRAPH_RERANKING_CANDIDATES=256
# CODEGRAPH_RERANKING_TOP_N=10

# --- Performance / concurrency ---
CODEGRAPH_WORKERS=4                          # caps Rayon threads
CODEGRAPH_MAX_CONCURRENT=4                   # concurrent embedding requests
# CODEGRAPH_SYMBOL_BATCH_SIZE=500
# CODEGRAPH_SYMBOL_MAX_CONCURRENT=4

# --- SurrealDB ---
# --- SurrealDB ---
# SurrealDB Configuration (for graph storage)
CODEGRAPH_SURREALDB_URL=ws://localhost:3004
CODEGRAPH_SURREALDB_NAMESPACE=main
CODEGRAPH_SURREALDB_DATABASE=codegraph
CODEGRAPH_SURREALDB_USERNAME=root
CODEGRAPH_SURREALDB_PASSWORD=root
#CODEGRAPH_USE_GRAPH_SCHEMA=true # Enable graph schema (experimental) if you setup the database with the codegraph_experimental.surql
#CODEGRAPH_GRAPH_DB_DATABASE=codegraph_experimental # experimental full graph schema has some bug that causes bad performance during indexing were investigating what causes this
CODEGRAPH_CHUNK_DB_BATCH_SIZE=512
CODEGRAPH_SURREAL_POOL_SIZE=2

# --- Provider endpoints & keys ---
CODEGRAPH_OLLAMA_URL=http://localhost:11434
# CODEGRAPH_LMSTUDIO_URL=http://localhost:1234
# OPENAI_API_KEY=sk-...
# ANTHROPIC_API_KEY=...
# JINA_API_KEY=...
# XAI_API_KEY=...
# OPENAI_API_BASE=https://api.openai.com/v1    # for openai-compatible providers

# --- LLM (for agent responses) ---
CODEGRAPH_LLM_PROVIDER=ollama                # ollama | openai | anthropic | openai-compatible | xai | lmstudio
CODEGRAPH_MODEL=qwen2.5-coder:14b
CODEGRAPH_CONTEXT_WINDOW=32768
#MCP_CODE_AGENT_MAX_OUTPUT_TOKENS=52000 # Hard-cap since f.ex. claude code doesn't support more than 64K and might crash on such outputs
# --- Server / daemon ---
CODEGRAPH_HTTP_HOST=127.0.0.1
CODEGRAPH_HTTP_PORT=3003
# CODEGRAPH_DAEMON_AUTO_START=true
# CODEGRAPH_WATCH=1                           # enable file watching when supported

# --- Logging ---
RUST_LOG=info

# --- AutoAgents ---
# CODEGRAPH_AUTOAGENTS_EXPERIMENTAL=1
# MCP_CODE_AGENT_MAX_OUTPUT_TOKENS=4096      # hard cap for autoagents responses

# --- Jina specific (when embedding provider=jina) ---
# JINA_API_KEY=...
#JINA_API_BASE=https://api.jina.ai/v1
#CODEGRAPH_EMBEDDING_MODEL=jina-embeddings-v4
#JINA_ENABLE_RERANKING=true
#JINA_RERANKING_MODEL=jina-reranker-v3
#JINA_MAX_TOKENS=256
#JINA_MAX_TEXTS=24
#JINA_REQUEST_DELAY_MS=100
#JINA_RERANKING_TOP_N=10
#JINA_LATE_CHUNKING=false # better embeddings with jina-embeddings-v4
#JINA_TRUNCATE=true
#JINA_API_TASK=code.passage
#CODEGRAPH_SYMBOL_BATCH_SIZE=64
#CODEGRAPH_SYMBOL_MAX_CONCURRENT=1
#JINA_REL_BATCH_SIZE=50
#JINA_REL_MAX_TEXTS=50

# --- Agent Configuration ---
CODEGRAPH_AGENT_ARCHITECTURE=rig             # rig (default) | react | lats | reflexion
                                             # Note: 'rig' backend automatically selects best sub-architecture
                                             # based on task complexity (LATS for deep tasks, ReAct for speed).

# Rig Agent Internal Tuning
# CODEGRAPH_AGENT_MAX_STEPS=8                # max tool calls per task (Hard cap: 10, default: 8)
# CODEGRAPH_AGENT_MEMORY_WINDOW=40           # number of turns to keep in history

# Dynamic Context Throttling (Rig Agent)
CODEGRAPH_CONTEXT_WINDOW=128000              # Set this to your model's actual token limit.
                                             # Rig Agent automatically downgrades tier (Detailed -> Terse)
                                             # when context usage > 80% to prevent overflow.

# AutoAgents specific (Legacy)
# CODEGRAPH_LATS_SELECTION_PROVIDER=openai
# CODEGRAPH_LATS_SELECTION_MODEL=gpt-5.1-codex-mini
# CODEGRAPH_LATS_EXPANSION_PROVIDER=openai
# CODEGRAPH_LATS_EXPANSION_MODEL=gpt-5.1-codex
# CODEGRAPH_LATS_EVALUATION_PROVIDER=openai
# CODEGRAPH_LATS_EVALUATION_MODEL=gpt-5.1
# CODEGRAPH_LATS_BEAM_WIDTH=3                # Number of best paths to keep (default: 3)
# CODEGRAPH_LATS_MAX_DEPTH=5

# MCP Server Max Output Tokens
MCP_CODE_AGENT_MAX_OUTPUT_TOKENS=58000       # Cap for final answers (Claude Code max ~64K)