Top AI Repos — open-source AI, indexed and scored

Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.

wujfeng712-ui/codex-bridge

Zero-dependency local proxy bridging Codex CLI to DeepSeek, Xiaomi MiMo, and OpenAI

51MITJavaScript

View on GitHub

Stars: 258
Forks: 39
Watchers: 2
Open issues: 11
Contributors: —

Star history, last 90 days— over 2 snapshots

2026-07-31 · 2582026-08-06 · 258

Star history, last 90 days
Date	Stars
2026-07-31	258
2026-08-06	258

Today

— stars today

This week

— stars this week

This month

— stars this month

Momentum

0.0

growth rate 0.00%/day

From the README

codex-bridge

A zero-dependency local proxy that lets Codex CLI talk to DeepSeek, Xiaomi MiMo, and OpenAI through a single base_url.

English · 简体中文 · 日本語 · 한국어 · Español

Codex CLI speaks the OpenAI Responses API. DeepSeek and MiMo speak Chat Completions. codex-bridge translates between them in both directions — streaming SSE, tool calls, and thinking-mode round-trips — so you can use any supported model inside Codex without patching the client.

Features

Multi-provider routing — DeepSeek / MiMo / OpenAI, auto-selected by model name
Bi-directional protocol translation — Responses API ↔ Chat Completions with streaming SSE bridge
Per-provider reasoning effort translation — Codex's none | minimal | low | medium | high | xhigh mapped to each upstream's native format
Thinking-mode tool-call round-trip — caches reasoning_content and replays it so DeepSeek's thinking mode survives multi-turn tool calls
Inbound auth gate — PROXY_AUTH_KEY / PROXY_KEYS with optional per-key provider locking
Session continuity — previous_response_id works across providers (LRU-bounded store)
Built-in web_fetch tool — bypasses sandbox restrictions for URL-heavy conversations
Tool-call circuit breaker — soft warning + hard tool stripping on runaway tool loops
Single-file, zero dependencies — one proxy.mjs (~2000 lines), no npm install

Quick Start

1. Configure

git clone https://github.com/wujfeng712-ui/codex-bridge.git
cd codex-bridge
cp env.example .env

Edit .env — at minimum:

PROXY_AUTH_KEY=sk-proxy-local-$(openssl rand -hex 24)   # generate one
DEEPSEEK_API_KEY=sk-...                                  # from platform.deepseek.com

2. Start the proxy

node --env-file=.env proxy.mjs

Need Node 18–19 or background mode? See Advanced Usage.

3. Point Codex CLI at the proxy

Edit ~/.codex/config.toml:

model = "deepseek-v4-flash"
model_provider = "local_proxy"

[model_providers.local_proxy]
name = "local_proxy"
base_url = "http://127.0.0.1:4000/v1"
wire_api = "responses"
requires_openai_auth = true

Set the auth key for Codex:

# ~/.codex/auth.json
{ "OPENAI_API_KEY": "<same PROXY_AUTH_KEY from .env>" }

Using CC Switch? Skip the manual edit — add a provider in the GUI instead. See Using with CC Switch.

Run codex — done.

Architecture

┌─────────────┐    Responses API    ┌──────────────┐
│  Codex CLI  │────────────────────▶│ codex-bridge │
│             │  Authorization:     │    :4000     │
└─────────────┘  Bearer <key>       └──────┬───────┘
                                           │  model-based routing
                   ┌───────────────────────┼────────────────────────┐
                   │                       │                        │
                   ▼                       ▼                        ▼
          ┌────────────────┐      ┌────────────────┐       ┌──────────────┐
          │   DeepSeek V4  │      │  Xiaomi MiMo   │       │    OpenAI    │
          │ Chat Complet.  │      │ Chat Complet.  │       │  Responses   │
          └────────────────┘      └────────────────┘       └──────────────┘

Configuration

All settings via environment variables (see env.example for full documentation):

Auth

Variable	Default	Description
`PROXY_AUTH_KEY`	—	Single inbound key (no provider lock)
`PROXY_KEYS`	—	Multi-key table: `<key>:<provider>,...` where provider ∈ `deepseek`/`mimo`/`openai`/`*`

Both empty = auth disabled (not recommended).

Upstream Providers

Variable	Default	Description
`DEEPSEEK_API_KEY`	—	DeepSeek upstream key
`DEEPSEEK_BASE_URL`	`https://api.deepseek.com/v1`	DeepSeek base URL
`DEEPSEEK_MODELS`	`deepseek-v4-pro,deepseek-v4-flash`	Models to advertise
`MIMO_API_KEY`	—	Xiaomi MiMo upstream key
`MIMO_BASE_URL`	`https://token-plan-cn.xiaomimimo.com/v1`	MiMo base URL
`MIMO_MODELS`	`mimo-v2.5-pro`	Models to advertise (must be lowercase)
`OPENAI_API_KEY`	—	OpenAI upstream key (opt-in)
`OPENAI_BASE_URL`	`https://api.openai.com/v1`	OpenAI base URL
`OPENAI_MODELS`	—	Explicit OpenAI model list
`OPENAI_MODEL_PREFIXES`	`gpt-,o1,o3,o4,codex-,chatgpt-`	Heuristic routing prefixes

Model Catalog

Variable	Default	Description
`MODEL_CATALOG_PATH`	—	Path to a `proxy-models.json` file. Overrides `*_MODELS` vars. Same file Codex reads via `model_catalog_json`

Tuning

Variable	Default	Description
`PROXY_PORT`	`4000`	Listen port
`DEFAULT_PROVIDER`	auto	Fallback when model is unknown
`LOG_LEVEL`	`info`	`silent` / `error` / `warn` / `info` / `debug`
`ACCESS_LOG`	on	Set `0` to suppress per-request access logs
`UPSTREAM_TIMEOUT_MS`	`120000`	Upstream request timeout
`STORE_TTL_MS`	`3600000`	Response store entry TTL
`STORE_MAX`	`500`	Response store LRU capacity
`GITHUB_TOKEN`	—	Optional; falls back to `gh auth token` lazily

Routing Rules

Each request is routed by model name, in priority order:

Exact match — model appears in DEEPSEEK_MODELS, MIMO_MODELS, or OPENAI_MODELS
Prefix heuristic — model starts with an OPENAI_MODEL_PREFIXES entry → OpenAI
Name hint — model contains deepseek or mimo → corresponding provider
Fallback — DEFAULT_PROVIDER, then first provider with a configured key

Reasoning Effort Translation

Codex effort	DeepSeek	MiMo	OpenAI
`none`	`thinking: {type: "disabled"}`	`thinking: {type: "disabled"}`	field removed
`minimal`	`reasoning_effort: "low"`	`reasoning_effort: "low"`	passthrough
`low` / `medium` / `high`	passthrough	passthrough	passthrough
`xhigh`	`reasoning_effort: "xhigh"`	clamped to `high`	clamped to `high`

Note: DeepSeek silently ignores enable_thinking: false. The proxy uses thinking: {type: "disabled"} instead.

Endpoints

Method	Path	Auth	Description
`GET`	`/health`	No	Health check
`GET`	`/v1/models`	Yes	Merged model list
`POST`	`/v1/responses`	Yes	Codex CLI main endpoint (Responses API)
`POST`	`/v1/chat/completions`	Yes	Direct Chat Completions passthrough
`GET`	`/cop?url=...`	Yes	URL fetch (Jina Reader / native HTTP)
`POST`	`/cop`	Yes	URL fetch with custom method/headers/body

Smoke Test

./scripts/smoke.sh                    # uses localhost:4000 by default
./scripts/smoke.sh http://host:4000   # custom target
MODEL=mimo-v2.5-pro ./scripts/smoke.sh  # test a different model

Runs 30 checks covering endpoints, input shapes, auth gate, streaming completion, effort translation, tool-call round-trips, and provider locking.

Using with CC Switch

CC Switch is a popular desktop app for managing provider profiles across AI CLI tools (Claude Code, Codex, Gemini CLI, etc.). You can use it to switch between codex-bridge and other providers with one click.

Setup

Open CC Switch → Codex tab → Add Provider
Fill in the provider fields:

Field Value

Name codex-bridge (or any label you prefer)

API Key Your PROXY_AUTH_KEY from .env

Base URL http://127.0.0.1:4000/v1
Click Enable to activate — CC Switch writes the key to ~/.codex/auth.json and updates config.toml automatically.

Field	Value
Name	`codex-bridge` (or any label you prefer)
API Key	Your `PROXY_AUTH_KEY` from `.env`
Base URL	`http://127.0.0.1:4000/v1`

Multi-provider switching

If you have multiple upstream keys (e.g., one for DeepSeek, one for MiMo), use PROXY_KEYS in .env to create per-provider inbound keys:

PROXY_KEYS=sk-deepseek-aaa:deepseek,sk-mimo-bbb:mimo,sk-all-ccc:*

Then create a separate CC Switch profile for each key — switching profiles swaps which upstream provider codex-bridge routes to.

CLI alternative

If you prefer the terminal, cc-switch-cli provides the same profile switching without a GUI:

# List profiles
cc-switch list

# Switch to the codex-bridge profile
cc-switch use codex-bridge

Tip: After switching profiles, restart your terminal (or run codex in a new shell) for the new auth to take effect.

Advanced Usage

Node 18–19 startup — --env-file was added in Node 20. On older versions:
```
set -a && source .env && set +a && node proxy.mjs
```

Background mode:

nohup node --env-file=.env proxy.mjs > /tmp/codex-bridge.log 2>&1 &

Multi-key provider locking — assign each inbound key to a specific provider for multi-profile setups. See env.example for the PROXY_KEYS format.
Model catalog single source of truth — point MODEL_CATALOG_PATH at the same JSON file Codex uses (model_catalog_json in config.toml) to keep model lists in sync automatically.

Troubleshooting

Symptom	Cause	Fix
`EADDRINUSE :4000`	Port already in use	`lsof -ti:4000 \| xargs kill` or change `PROXY_PORT` in `.env`
`401 Unauthorized`	Auth key mismatch	Ensure `OPENAI_API_KEY` in `~/.codex/auth.json` matches `PROXY_AUTH_KEY` in `.env`
`--env-file: not recognized`	Node.js < 20	Use `set -a && source .env && set +a && node proxy.mjs`
Upstream timeout	Slow provider response	Increase `UPSTREAM_TIMEOUT_MS` in `.env` (default 120 000 ms)
Model not found	Model not in any `*_MODELS` list	Add it to `DEEPSEEK_MODELS` / `MIMO_MODELS` / `OPENAI_MODELS`, or use `MODEL_CATALOG_PATH`

Requirements

Node.js 18+
macOS / Linux / Windows
At least one upstream API key (DeepSeek, MiMo, or OpenAI)

License

MIT — see LICENSE.

Rendered by GitHub from the current README

Read on GitHub

Top contributors

No contributor data yet — it is fetched during the contributor sync pass.

Quality score

Would you bet a product on this? Bounded 0–100 and slow moving.

51/10051

Maintenance: 45%; How recently the default branch was pushed to
Releases: 25%; Recency and cadence of tagged releases
Community: 20%; Contributor breadth and bus factor
Issue backlog: 60%; Open issues measured against the size of the audience
Documentation: 83%; README depth, plus a homepage and topics
Licensing: 100%; How freely the code can be adopted

What to know before adopting

No tagged releases
Usable in commercial products with attribution.

Details

Created: 3 months ago
Last push: 3 months ago
Last release: none
Release tag: —
Releases in the last year: 0
Default branch: main
Size: 70 KB
Owner type: User
Owner location: —
Discovered via: phrase:codex cli
First indexed: 7 days ago
Last synced: 3 hours ago

wujfeng712-ui/codex-bridge

Zero-dependency local proxy bridging Codex CLI to DeepSeek, Xiaomi MiMo, and OpenAI

51MITJavaScript

View on GitHub

Stars: 258
Forks: 39
Watchers: 2
Open issues: 11
Contributors: —

Star history, last 90 days— over 2 snapshots

2026-07-31 · 2582026-08-06 · 258

Star history, last 90 days
Date	Stars
2026-07-31	258
2026-08-06	258

Today

— stars today

This week

— stars this week

This month

— stars this month

Momentum

0.0

growth rate 0.00%/day

From the README

codex-bridge

A zero-dependency local proxy that lets Codex CLI talk to DeepSeek, Xiaomi MiMo, and OpenAI through a single base_url.

English · 简体中文 · 日本語 · 한국어 · Español

Features

Multi-provider routing — DeepSeek / MiMo / OpenAI, auto-selected by model name
Bi-directional protocol translation — Responses API ↔ Chat Completions with streaming SSE bridge
Per-provider reasoning effort translation — Codex's none | minimal | low | medium | high | xhigh mapped to each upstream's native format
Thinking-mode tool-call round-trip — caches reasoning_content and replays it so DeepSeek's thinking mode survives multi-turn tool calls
Inbound auth gate — PROXY_AUTH_KEY / PROXY_KEYS with optional per-key provider locking
Session continuity — previous_response_id works across providers (LRU-bounded store)
Built-in web_fetch tool — bypasses sandbox restrictions for URL-heavy conversations
Tool-call circuit breaker — soft warning + hard tool stripping on runaway tool loops
Single-file, zero dependencies — one proxy.mjs (~2000 lines), no npm install

Quick Start

1. Configure

git clone https://github.com/wujfeng712-ui/codex-bridge.git
cd codex-bridge
cp env.example .env

Edit .env — at minimum:

PROXY_AUTH_KEY=sk-proxy-local-$(openssl rand -hex 24)   # generate one
DEEPSEEK_API_KEY=sk-...                                  # from platform.deepseek.com

2. Start the proxy

node --env-file=.env proxy.mjs

Need Node 18–19 or background mode? See Advanced Usage.

3. Point Codex CLI at the proxy

Edit ~/.codex/config.toml:

model = "deepseek-v4-flash"
model_provider = "local_proxy"

[model_providers.local_proxy]
name = "local_proxy"
base_url = "http://127.0.0.1:4000/v1"
wire_api = "responses"
requires_openai_auth = true

Set the auth key for Codex:

# ~/.codex/auth.json
{ "OPENAI_API_KEY": "<same PROXY_AUTH_KEY from .env>" }

Using CC Switch? Skip the manual edit — add a provider in the GUI instead. See Using with CC Switch.

Run codex — done.

Architecture

┌─────────────┐    Responses API    ┌──────────────┐
│  Codex CLI  │────────────────────▶│ codex-bridge │
│             │  Authorization:     │    :4000     │
└─────────────┘  Bearer <key>       └──────┬───────┘
                                           │  model-based routing
                   ┌───────────────────────┼────────────────────────┐
                   │                       │                        │
                   ▼                       ▼                        ▼
          ┌────────────────┐      ┌────────────────┐       ┌──────────────┐
          │   DeepSeek V4  │      │  Xiaomi MiMo   │       │    OpenAI    │
          │ Chat Complet.  │      │ Chat Complet.  │       │  Responses   │
          └────────────────┘      └────────────────┘       └──────────────┘

Configuration

All settings via environment variables (see env.example for full documentation):

Auth

Variable	Default	Description
`PROXY_AUTH_KEY`	—	Single inbound key (no provider lock)
`PROXY_KEYS`	—	Multi-key table: `<key>:<provider>,...` where provider ∈ `deepseek`/`mimo`/`openai`/`*`

Both empty = auth disabled (not recommended).

Upstream Providers

Variable	Default	Description
`DEEPSEEK_API_KEY`	—	DeepSeek upstream key
`DEEPSEEK_BASE_URL`	`https://api.deepseek.com/v1`	DeepSeek base URL
`DEEPSEEK_MODELS`	`deepseek-v4-pro,deepseek-v4-flash`	Models to advertise
`MIMO_API_KEY`	—	Xiaomi MiMo upstream key
`MIMO_BASE_URL`	`https://token-plan-cn.xiaomimimo.com/v1`	MiMo base URL
`MIMO_MODELS`	`mimo-v2.5-pro`	Models to advertise (must be lowercase)
`OPENAI_API_KEY`	—	OpenAI upstream key (opt-in)
`OPENAI_BASE_URL`	`https://api.openai.com/v1`	OpenAI base URL
`OPENAI_MODELS`	—	Explicit OpenAI model list
`OPENAI_MODEL_PREFIXES`	`gpt-,o1,o3,o4,codex-,chatgpt-`	Heuristic routing prefixes

Model Catalog

Variable	Default	Description
`MODEL_CATALOG_PATH`	—	Path to a `proxy-models.json` file. Overrides `*_MODELS` vars. Same file Codex reads via `model_catalog_json`

Tuning

Variable	Default	Description
`PROXY_PORT`	`4000`	Listen port
`DEFAULT_PROVIDER`	auto	Fallback when model is unknown
`LOG_LEVEL`	`info`	`silent` / `error` / `warn` / `info` / `debug`
`ACCESS_LOG`	on	Set `0` to suppress per-request access logs
`UPSTREAM_TIMEOUT_MS`	`120000`	Upstream request timeout
`STORE_TTL_MS`	`3600000`	Response store entry TTL
`STORE_MAX`	`500`	Response store LRU capacity
`GITHUB_TOKEN`	—	Optional; falls back to `gh auth token` lazily

Routing Rules

Each request is routed by model name, in priority order:

Exact match — model appears in DEEPSEEK_MODELS, MIMO_MODELS, or OPENAI_MODELS
Prefix heuristic — model starts with an OPENAI_MODEL_PREFIXES entry → OpenAI
Name hint — model contains deepseek or mimo → corresponding provider
Fallback — DEFAULT_PROVIDER, then first provider with a configured key

Reasoning Effort Translation

Codex effort	DeepSeek	MiMo	OpenAI
`none`	`thinking: {type: "disabled"}`	`thinking: {type: "disabled"}`	field removed
`minimal`	`reasoning_effort: "low"`	`reasoning_effort: "low"`	passthrough
`low` / `medium` / `high`	passthrough	passthrough	passthrough
`xhigh`	`reasoning_effort: "xhigh"`	clamped to `high`	clamped to `high`

Note: DeepSeek silently ignores enable_thinking: false. The proxy uses thinking: {type: "disabled"} instead.

Endpoints

Method	Path	Auth	Description
`GET`	`/health`	No	Health check
`GET`	`/v1/models`	Yes	Merged model list
`POST`	`/v1/responses`	Yes	Codex CLI main endpoint (Responses API)
`POST`	`/v1/chat/completions`	Yes	Direct Chat Completions passthrough
`GET`	`/cop?url=...`	Yes	URL fetch (Jina Reader / native HTTP)
`POST`	`/cop`	Yes	URL fetch with custom method/headers/body

Smoke Test

./scripts/smoke.sh                    # uses localhost:4000 by default
./scripts/smoke.sh http://host:4000   # custom target
MODEL=mimo-v2.5-pro ./scripts/smoke.sh  # test a different model

Runs 30 checks covering endpoints, input shapes, auth gate, streaming completion, effort translation, tool-call round-trips, and provider locking.

Using with CC Switch

Setup

Open CC Switch → Codex tab → Add Provider
Fill in the provider fields:

Field Value

Name codex-bridge (or any label you prefer)

API Key Your PROXY_AUTH_KEY from .env

Base URL http://127.0.0.1:4000/v1
Click Enable to activate — CC Switch writes the key to ~/.codex/auth.json and updates config.toml automatically.

Field	Value
Name	`codex-bridge` (or any label you prefer)
API Key	Your `PROXY_AUTH_KEY` from `.env`
Base URL	`http://127.0.0.1:4000/v1`

Multi-provider switching

If you have multiple upstream keys (e.g., one for DeepSeek, one for MiMo), use PROXY_KEYS in .env to create per-provider inbound keys:

PROXY_KEYS=sk-deepseek-aaa:deepseek,sk-mimo-bbb:mimo,sk-all-ccc:*

Then create a separate CC Switch profile for each key — switching profiles swaps which upstream provider codex-bridge routes to.

CLI alternative

If you prefer the terminal, cc-switch-cli provides the same profile switching without a GUI:

# List profiles
cc-switch list

# Switch to the codex-bridge profile
cc-switch use codex-bridge

Tip: After switching profiles, restart your terminal (or run codex in a new shell) for the new auth to take effect.

Advanced Usage

Node 18–19 startup — --env-file was added in Node 20. On older versions:
```
set -a && source .env && set +a && node proxy.mjs
```

Background mode:

nohup node --env-file=.env proxy.mjs > /tmp/codex-bridge.log 2>&1 &

Multi-key provider locking — assign each inbound key to a specific provider for multi-profile setups. See env.example for the PROXY_KEYS format.
Model catalog single source of truth — point MODEL_CATALOG_PATH at the same JSON file Codex uses (model_catalog_json in config.toml) to keep model lists in sync automatically.

Troubleshooting

Symptom	Cause	Fix
`EADDRINUSE :4000`	Port already in use	`lsof -ti:4000 \| xargs kill` or change `PROXY_PORT` in `.env`
`401 Unauthorized`	Auth key mismatch	Ensure `OPENAI_API_KEY` in `~/.codex/auth.json` matches `PROXY_AUTH_KEY` in `.env`
`--env-file: not recognized`	Node.js < 20	Use `set -a && source .env && set +a && node proxy.mjs`
Upstream timeout	Slow provider response	Increase `UPSTREAM_TIMEOUT_MS` in `.env` (default 120 000 ms)
Model not found	Model not in any `*_MODELS` list	Add it to `DEEPSEEK_MODELS` / `MIMO_MODELS` / `OPENAI_MODELS`, or use `MODEL_CATALOG_PATH`

Requirements

Node.js 18+
macOS / Linux / Windows
At least one upstream API key (DeepSeek, MiMo, or OpenAI)

License

MIT — see LICENSE.

Rendered by GitHub from the current README

Read on GitHub

Top contributors

No contributor data yet — it is fetched during the contributor sync pass.

Quality score

Would you bet a product on this? Bounded 0–100 and slow moving.

51/10051

Maintenance: 45%; How recently the default branch was pushed to
Releases: 25%; Recency and cadence of tagged releases
Community: 20%; Contributor breadth and bus factor
Issue backlog: 60%; Open issues measured against the size of the audience
Documentation: 83%; README depth, plus a homepage and topics
Licensing: 100%; How freely the code can be adopted

What to know before adopting

No tagged releases
Usable in commercial products with attribution.

Details

Created: 3 months ago
Last push: 3 months ago
Last release: none
Release tag: —
Releases in the last year: 0
Default branch: main
Size: 70 KB
Owner type: User
Owner location: —
Discovered via: phrase:codex cli
First indexed: 7 days ago
Last synced: 3 hours ago

Top AI Repos — open-source AI, indexed and scored

Trend

From the README

codex-bridge

Features

Quick Start

1. Configure

2. Start the proxy

3. Point Codex CLI at the proxy

Architecture

Configuration

Auth

Upstream Providers

Model Catalog

Tuning

Routing Rules

Reasoning Effort Translation

Endpoints

Smoke Test

Using with CC Switch

Setup

Multi-provider switching

CLI alternative

Advanced Usage

Troubleshooting

Requirements

License

Top contributors

Quality score

What to know before adopting

Categories

Details

Trend

From the README

codex-bridge

Features

Quick Start

1. Configure

2. Start the proxy

3. Point Codex CLI at the proxy

Architecture

Configuration

Auth

Upstream Providers

Model Catalog

Tuning

Routing Rules

Reasoning Effort Translation

Endpoints

Smoke Test

Using with CC Switch

Setup

Multi-provider switching

CLI alternative

Advanced Usage

Troubleshooting

Requirements

License

Top contributors

Quality score

What to know before adopting

Categories

Details