ferryllm 0.1.4

Universal LLM protocol middleware for OpenAI, Anthropic, Claude Code, and OpenAI-compatible backends.
Documentation 
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

[server]
listen = "0.0.0.0:3000"
request_timeout_secs = 120
body_limit_mb = 32
# Optional. Uncomment to cap in-flight requests.
# max_concurrent_requests = 128
# Optional. Uncomment to cap total requests per minute.
# rate_limit_per_minute = 600
# Optional non-streaming upstream resilience. Streaming requests are not retried.
# retry_attempts = 2
# retry_backoff_ms = 100
# circuit_breaker_failures = 5
# circuit_breaker_cooldown_secs = 30

[logging]
level = "info"
format = "text"

[auth]
enabled = false
# api_keys_env = "FERRYLLM_API_KEYS"
# Optional per-client caps, keyed by the authenticated API key.
# per_key_rate_limit_per_minute = 120
# per_key_max_concurrent_requests = 8

[metrics]
enabled = true

[prompt_cache]
# LiteLLM-style automatic Anthropic cache breakpoint injection.
# This preserves client-provided cache_control and adds ephemeral breakpoints
# on stable system/tools/last user blocks when they are missing.
auto_inject_anthropic_cache_control = true
cache_system = true
cache_tools = true
cache_last_user_message = true
# OpenAI-compatible prompt cache routing key. Keep it stable; do not include
# request-specific conversation text in this value.
openai_prompt_cache_key = "ferryllm"
# Optional, if the upstream accepts it.
# openai_prompt_cache_retention = "24h"
# Safe long-running diagnostic: logs request structure, lengths, and hashes
# without logging prompt text or tool-result bodies.
debug_log_request_shape = true
# Claude Code may place a volatile line near the beginning of system text.
# Move the full line intersecting this byte range into a later user context
# block so stable system instructions remain first for provider prompt caches.
relocate_system_prefix_range = "0..1"
# Temporary investigation only: this prints the relocated prompt text.
# Disable after confirming the moved line.
log_relocated_system_text = false
# Strip transport metadata lines from system before forwarding.
strip_system_line_prefixes = ["x-anthropic-billing-header:"]

[[providers]]
name = "codexapis"
type = "openai"
base_url = "https://codexapis.com"
api_key_env = "CODX_API_KEY"

[[routes]]
match = "cc-gpt55"
match_type = "exact"
provider = "codexapis"
rewrite_model = "gpt-5.4"

[[routes]]
match = "claude-"
provider = "codexapis"
rewrite_model = "gpt-5.4"

[[routes]]
match = "gpt-"
provider = "codexapis"

[[routes]]
match = "grok-"
provider = "codexapis"

[[routes]]
match = "*"
provider = "codexapis"
rewrite_model = "gpt-5.4"