Cargo.toml

[workspace]
resolver = "2"
members = [
    "crates/infernum",           # Main CLI/binary
    "crates/infernum-core",      # Shared types and traits
    "crates/abaddon",            # Inference engine
    "crates/malphas",            # Orchestration layer
    "crates/stolas",             # Knowledge/RAG engine
    "crates/beleth",             # Agent framework
    "crates/asmodeus",           # Fine-tuning/adaptation
    "crates/dantalion",          # Observability
    "crates/infernum-server",    # HTTP API server
    "crates/grimoire-loader",    # Grimoire integration
    "crates/paimon",             # LLM Studio
    "crates/legion",             # Holographic agent swarm
    "crates/arbiter",            # Unified GPU coordination
    "crates/test-utils",         # Testing utilities
]

[workspace.package]
version = "0.2.0-rc.2"
edition = "2021"
rust-version = "1.91"
authors = ["Daemoniorum Engineering <engineering@daemoniorum.com>"]
license = "MIT OR Apache-2.0"
repository = "https://github.com/Daemoniorum-LLC/infernum"
homepage = "https://infernum.daemoniorum.com"
readme = "README.md"
description = "Local LLM inference framework"
keywords = ["llm", "inference", "ai", "ml", "gpu"]
categories = ["science", "api-bindings", "command-line-utilities"]

[workspace.dependencies]
# Async runtime
tokio = { version = "1.48", features = ["full"] }
tokio-stream = { version = "0.1", features = ["sync"] }
futures = "0.3"
async-trait = "0.1"

# Serialization
serde = { version = "1.0", features = ["derive"] }
serde_json = "1.0"
serde_yaml = "0.9"
bincode = "1.3"
toml = "0.9"

# HTTP/API
axum = { version = "0.8", features = ["macros", "ws"] }
axum-server = { version = "0.8", features = ["tls-rustls"] }
tower = { version = "0.5", features = ["full"] }
tower-http = { version = "0.6", features = ["cors", "trace", "compression-gzip", "timeout"] }
reqwest = { version = "0.13", features = ["json", "stream"] }
hyper = { version = "1.6", features = ["full"] }
rustls-pemfile = "2.2"

# ML/Tensor operations
# Note: default-features = false to avoid pulling in CUDA unconditionally
# GPU features are enabled via crate feature flags (cuda, metal)
candle-core = { version = "0.9", default-features = false }
candle-nn = { version = "0.9", default-features = false }
candle-transformers = { version = "0.9", default-features = false }
candle-flash-attn = "0.9"
safetensors = "0.7"
tokenizers = "0.22"
hf-hub = "0.4"
half = "2.4"

# CUDA runtime (for GPU queries) - using 0.12 for API stability
# cuda-version-from-build-system auto-detects CUDA version at build time
# cublas for GPU-accelerated matrix operations (used in HoloTensor SVD)
cudarc = { version = "0.19", features = ["cuda-version-from-build-system", "f16", "cublas"] }

# Vector storage
lance = "2.0"
lancedb = "0.26"
arrow-array = "54"
arrow-schema = "54"

# gRPC
tonic = "0.14"
tonic-build = "0.14"
prost = "0.14"
prost-types = "0.14"

# Observability
tracing = "0.1"
tracing-subscriber = { version = "0.3", features = ["env-filter", "json"] }
opentelemetry = "0.31"
opentelemetry-otlp = { version = "0.31", features = ["grpc-tonic"] }
opentelemetry_sdk = { version = "0.31", features = ["rt-tokio"] }
metrics = "0.24"
metrics-exporter-prometheus = "0.18"

# Error handling
thiserror = "2.0"
anyhow = "1.0"
color-eyre = "0.6"

# CLI
clap = { version = "4.5", features = ["derive", "env"] }
dialoguer = "0.12"
indicatif = "0.18"
console = "0.16"

# Configuration
figment = { version = "0.10", features = ["toml", "env", "yaml"] }
config = "0.14"

# File watching (unified on 8.0)
notify = { version = "8.0", default-features = false, features = ["macos_fsevent"] }

# Utilities
uuid = { version = "1.20", features = ["v4", "serde"] }
chrono = { version = "0.4", features = ["serde"] }
rand = "0.9"
hex = "0.4"
dashmap = "6.1"
parking_lot = "0.12"
bytes = "1.11"
memmap2 = "0.9"
rayon = "1.10"
crossbeam = "0.8"
num_cpus = "1.16"
humansize = "2.1"
regex = "1.12"

# Syntax parsing (14 languages for code analysis)
tree-sitter = "0.24"
tree-sitter-python = "0.23"
tree-sitter-rust = "0.23"
tree-sitter-typescript = "0.23"
tree-sitter-go = "0.23"
tree-sitter-json = "0.24"
tree-sitter-yaml = "0.7"
tree-sitter-css = "0.23"
tree-sitter-bash = "0.23"
tree-sitter-html = "0.23"
tree-sitter-java = "0.23"
tree-sitter-c = "0.23"
tree-sitter-cpp = "0.23"
tree-sitter-c-sharp = "0.23"

# Testing
criterion = "0.6"
proptest = "1.10"
test-log = "0.2"
mockall = "0.14"
wiremock = "0.6"

# Compression (Haagenti - compressed tensor storage)
# Published on crates.io
haagenti = "0.1"
haagenti-core = "0.1"
haagenti-cuda = { version = "0.1", features = ["cufft"] }
haagenti-hct = "0.1"
# haagenti-neural = "0.1"  # Not yet published

# Table formatting
comfy-table = "7.2"

# Audit chain (Moloch) - published to crates.io
moloch-core = "0.1.0"
moloch-holocrypt = { version = "0.1.0", default-features = false, features = ["encryption"] }

# Cryptography (Arcanum - Phase 4 secure storage)
# Using crates.io versions for dogfooding
arcanum-symmetric = { version = "0.1.2", features = ["chacha20"] }
arcanum-hash = { version = "0.1.2", features = ["argon2", "blake3"] }
arcanum-core = "0.1.2"

[workspace.lints.rust]
# Allow unsafe - used intentionally for FFI and perf-critical sections
unsafe_code = "allow"
missing_docs = "warn"

[workspace.lints.clippy]
# Lint groups with lower priority so individual lints can override
all = { level = "warn", priority = -1 }
pedantic = { level = "warn", priority = -1 }
nursery = { level = "warn", priority = -1 }

# Safety lints - reviewed case-by-case, allowed for release
# These usages have been audited in the codebase
unwrap_used = "allow"
expect_used = "allow"
panic = "allow"
todo = "warn"
unimplemented = "warn"

# Allow high-volume pedantic lints that are cosmetic/stylistic
# These can be addressed incrementally post-release
must_use_candidate = "allow"
missing_errors_doc = "allow"
missing_panics_doc = "allow"
module_name_repetitions = "allow"
similar_names = "allow"
doc_markdown = "allow"
items_after_statements = "allow"
needless_raw_string_hashes = "allow"
match_same_arms = "allow"
redundant_else = "allow"
too_many_lines = "allow"
cognitive_complexity = "allow"
significant_drop_tightening = "allow"
cast_possible_truncation = "allow"
cast_sign_loss = "allow"
cast_precision_loss = "allow"
cast_lossless = "allow"
wildcard_imports = "allow"
struct_excessive_bools = "allow"
fn_params_excessive_bools = "allow"
single_match_else = "allow"
unnested_or_patterns = "allow"
if_not_else = "allow"
option_if_let_else = "allow"
manual_let_else = "allow"
trivially_copy_pass_by_ref = "allow"
needless_pass_by_value = "allow"
unnecessary_wraps = "allow"
unused_async = "allow"
return_self_not_must_use = "allow"
used_underscore_binding = "allow"
iter_without_into_iter = "allow"
implicit_hasher = "allow"
uninlined_format_args = "allow"
missing_const_for_fn = "allow"
use_self = "allow"
format_push_string = "allow"
map_unwrap_or = "allow"
unused_self = "allow"
redundant_closure_for_method_calls = "allow"
suboptimal_flops = "allow"
unreadable_literal = "allow"
io_other_error = "allow"
unnecessary_map_or = "allow"
future_not_send = "allow"
unnecessary_literal_bound = "allow"
derive_partial_eq_without_eq = "allow"
cast_possible_wrap = "allow"
or_fun_call = "allow"
derivable_impls = "allow"
redundant_clone = "allow"
assigning_clones = "allow"
manual_div_ceil = "allow"
ref_as_ptr = "allow"
large_futures = "allow"
missing_fields_in_debug = "allow"
branches_sharing_code = "allow"
equatable_if_let = "allow"
redundant_pub_crate = "allow"
option_option = "allow"
empty_docs = "allow"
transmute_ptr_to_ptr = "allow"
ignored_unit_patterns = "allow"
cloned_instead_of_copied = "allow"
case_sensitive_file_extension_comparisons = "allow"
useless_format = "allow"
struct_field_names = "allow"
significant_drop_in_scrutinee = "allow"
redundant_closure = "allow"
unnecessary_cast = "allow"
field_reassign_with_default = "allow"
too_many_arguments = "allow"
print_literal = "allow"
needless_borrows_for_generic_args = "allow"
manual_is_multiple_of = "allow"
explicit_iter_loop = "allow"
collapsible_if = "allow"
await_holding_lock = "allow"
unnecessary_lazy_evaluations = "allow"
unchecked_time_subtraction = "allow"
needless_range_loop = "allow"
single_char_pattern = "allow"
let_and_return = "allow"
explicit_into_iter_loop = "allow"
collection_is_never_read = "allow"
string_lit_as_bytes = "allow"
useless_conversion = "allow"
len_without_is_empty = "allow"
useless_vec = "allow"
borrow_as_ptr = "allow"
debug_assert_with_mut_call = "allow"
non_std_lazy_statics = "allow"
needless_continue = "allow"
needless_collect = "allow"
manual_midpoint = "allow"
implicit_clone = "allow"
clone_on_copy = "allow"
wildcard_in_or_patterns = "allow"
useless_let_if_seq = "allow"
unnecessary_debug_formatting = "allow"
type_complexity = "allow"
redundant_pattern_matching = "allow"
needless_pass_by_ref_mut = "allow"
needless_borrow = "allow"
many_single_char_names = "allow"
default_trait_access = "allow"
useless_asref = "allow"
unwrap_or_default = "allow"
should_implement_trait = "allow"
ptr_arg = "allow"
naive_bytecount = "allow"
match_wildcard_for_single_variants = "allow"
map_identity = "allow"
map_entry = "allow"
manual_strip = "allow"
manual_pattern_char_comparison = "allow"
get_first = "allow"
for_kv_map = "allow"
float_cmp = "allow"
elidable_lifetime_names = "allow"
bool_to_int_with_if = "allow"
wrong_self_convention = "allow"
vec_init_then_push = "allow"
unnecessary_struct_initialization = "allow"
to_string_in_format_args = "allow"
semicolon_if_nothing_returned = "allow"
result_large_err = "allow"
ref_option = "allow"
ptr_as_ptr = "allow"
no_effect_underscore_binding = "allow"
new_ret_no_self = "allow"
needless_question_mark = "allow"
needless_for_each = "allow"
mixed_attributes_style = "allow"
missing_const_for_thread_local = "allow"
map_flatten = "allow"
manual_non_exhaustive = "allow"
manual_find = "allow"
literal_string_with_formatting_args = "allow"
let_unit_value = "allow"
large_enum_variant = "allow"
iter_with_drain = "allow"
imprecise_flops = "allow"
if_same_then_else = "allow"
double_ended_iterator_last = "allow"
doc_link_with_quotes = "allow"

[profile.release]
lto = "thin"
codegen-units = 1
opt-level = 3
strip = "symbols"

[profile.release-with-debug]
inherits = "release"
debug = true
strip = "none"

[profile.dev]
opt-level = 1

[profile.dev.package."*"]
opt-level = 3