Repository navigation
Expand file tree
/
Copy pathNOTICE
More file actions
69 lines (57 loc) · 4.17 KB
/
Copy pathNOTICE
File metadata and controls
69 lines (57 loc) · 4.17 KB
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
teellm contains code derived from BerriAI/LiteLLM.
Copyright (c) 2023 Berri AI. Licensed under the MIT License; see LICENSE.
Upstream: https://github.com/BerriAI/litellm
Revision: 98465fe2ba4fc26ab9e6ab366c7f90f8e4d9c685
The SSE decoder in src/sse.rs is extracted and modified from
litellm-rust/crates/framer/src/sse.rs. The streaming transport follows
litellm-rust/crates/framer/src/framed.rs. Retired Claude request/response and usage
normalization were adapted from litellm-rust/crates/llms/src/anthropic/chat/
transformation.rs and litellm-rust/crates/core-utils/src/core_helpers.rs.
Retired compatible-provider parameter mapping followed litellm-rust/crates/llms/src/
openai_like/chat/transformation.rs.
Modifications remove the surrounding workspace dependencies,
restrict providers to OpenAI, Claude and Grok, introduce a standalone HTTP client, and implement
native protocol forwarding. The former Chat-to-Claude and stream conversion implementation has been removed.
The standalone HTTP server, authentication, configuration, multi-account
scheduler, retry/fallback logic, circuit breakers, limits and in-memory usage collection
are newly implemented in Rust for this project.
OAuth ports added from the same LiteLLM revision:
- litellm/llms/chatgpt/{authenticator,common_utils}.py and responses/transformation.py
- litellm/llms/xai/oauth.py and its Chat/Responses integration
- litellm-rust/crates/llms/src/anthropic/common_utils.rs (Rust extraction)
- litellm/completion_extras/litellm_responses_transformation/transformation.py
ChatGPT instruction injection and its instruction asset were removed; Agent instructions pass through unchanged.
Selected upstream authentication test cases were adapted into Rust tests.
UPSTREAM.json records ported functions, integration changes and validation.
Automatic model discovery is a new Rust integration using provider REST schemas,
LiteLLM model-list/OAuth header behavior and
the Codex model endpoint/query schema. It does not copy Codex implementation code
or a static model catalog. Pinned protocol references are recorded in UPSTREAM.json.
In-memory state/credential injection/CLI integration is new teellm code, not an upstream LiteLLM port.
Online administration, validated configuration generations, provider compatibility
profiles, account health, session affinity, shared account limits,
bounded resource handling, Router checkpoint/restore and operational diagnostics
are new teellm Rust integration code. OAuth algorithms retain the provenance above.
Declarative discovery header/query configuration is new teellm code.
Generation body rewrites and JSON/form conversion are forbidden. It includes no
script interpreter or replacement OAuth/streaming protocol implementation.
Official-host enforcement, raw model routing with separate pool selection, and
native reasoning-parameter preservation/protection are new teellm Rust integration.
The private teerouter-egress component links unmodified Xray-core v1.260327.0
(upstream release v26.3.27), https://github.com/XTLS/Xray-core, under MPL-2.0.
Source is available at https://github.com/XTLS/Xray-core/tree/v26.3.27.
The helper and Rust pipe/SOCKS lifecycle integration are new teerouter code.
Exact transitive versions and checksums are recorded in egress/go.mod and go.sum.
Dependency copyright and license notices must be preserved when redistributing
the helper and its dependencies; the helper is part of the Router trust boundary.
Error/interruption-only text estimation uses unmodified tiktoken-rs 0.12.1
(MIT), https://github.com/zurawiki/tiktoken-rs,
revision 72a5a800651a1cea1ad609292446ba02c80e3bcd. The bundled tokenizer
assets originate from OpenAI tiktoken, https://github.com/openai/tiktoken
(MIT). Their license texts are in LICENSES/tiktoken-rs-LICENSE.txt and
LICENSES/tiktoken-LICENSE.txt. The bounded evidence collection and estimate
billing policy are new teerouter code; estimated counts are not provider usage.
The optional accounting journal uses unmodified rusqlite 0.37.0 and
libsqlite3-sys 0.35.0 under the MIT License (LICENSES/rusqlite-LICENSE.txt and
LICENSES/libsqlite3-sys-LICENSE.txt), including the bundled public-domain SQLite
engine. Sealed-row storage, recovery and receipt scheduling are new teerouter code.