Agent Node

CyberCode can expose configured models and smart routes to other agents through a scoped node supporting both OpenAI Chat Completions and Anthropic Messages. Provider credentials remain inside CyberCode; the receiving agent only gets a separate node key.

Prepare the node

1. Open Models & Routing → Node.
2. Create a separate node API key for each user or agent. The complete key is shown only once.
3. Click the key's row in the table, select the models and routes it may access, then choose its default target for auto.
4. Save, then use the Connection configuration builder to choose a protocol and target. CyberCode generates a card with the Base URL, full endpoint, Model, and node key, with per-field and copy-all actions.

You can also configure a standalone TUI directly:

/node start
/node allow all
/node status

/node start starts CyberCode's built-in local runtime on demand and creates a cc_... key when none exists. The complete key is shown once. Run /node for an interactive default-target picker, or use /node default <target-id> in scripts. Use /node limit <count> for a monthly request limit, /node rotate to replace the key, /node stop to pause the node, and /node revoke to remove the key. /agent-node and /gateway are aliases.

Desktop-managed sessions

When the TUI was launched by the CyberCode desktop app, the desktop local server owns the node. Manage it in desktop settings so two processes do not modify the same key and port.

Complete fill-in example

The following example connects a “CI coding agent.” node.example.com is a documentation-only placeholder domain and cc_REPLACE_WITH_YOUR_NODE_KEY is an invalid redacted key. Replace them with the address and complete key shown by your own node.

Add an OpenAI Compatible provider in the receiving agent and fill only four fields:

Receiving field	Example value
Protocol	OpenAI Chat Completions
Base URL	https://node.example.com/v1
API key	cc_REPLACE_WITH_YOUR_NODE_KEY
Model	auto

For normal use, keep Model set to auto. No other advanced field is required. If the receiving agent also asks for a provider Name, use any recognizable local label, such as CyberCode Work Node. It does not control routing; provider aliases in Model distinguish the same model served by different upstream providers.

Use the same values for a connection test:

curl https://node.example.com/v1/chat/completions \
  -H "Authorization: Bearer cc_REPLACE_WITH_YOUR_NODE_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "auto",
    "messages": [
      {"role": "user", "content": "Reply only with: node connected"}
    ]
  }'

Node-side policy in CyberCode

Under Node → Access keys → Target policy, configure this key using the names shown in the UI:

Policy	Example
Key name	CI coding agent
Allowed access	Coding route, Kimi K2.6
Default target	Coding route
Monthly request limit	5000

CyberCode enforces these settings while the receiving agent continues to use Model=auto. Consult the later advanced Model section and copy an exact ID only when you intentionally need to pin one model or route.

Manage keys for multiple users

Give each user, device, or external agent its own key instead of sharing one value. This keeps scope, monthly quota, and usage separate, and lets you disable one compromised key without interrupting everyone else.

1. Click Add key and enter a recognizable name such as “Alice”, “CI runner”, or “Telegram bot”.
2. Copy the complete cc_... key immediately. CyberCode stores only its hash, so the complete value cannot be recovered after the app is refreshed or closed.
3. Click that key's row. The Target policy, monthly limit, and Connection configuration builder below now apply only to the selected key.
4. Select its allowed models and routes, set its auto target and monthly request limit, then save.
5. Choose a protocol and target in the builder, then enter the generated URL, Model, and complete key in the matching agent.

Actions in the key table:

Action	Result
Rename	Changes only the management label; the key value and connection keep working
Copy	Available only while the complete value remains in memory after creation or rotation
Rotate	Replaces only this key; the old value stops immediately while scope, quota, and current usage remain
Revoke	Blocks only this user; all other keys and the node continue working

If the key is already masked, choose Rotate key and update that agent with the new value. Removing the final key disables the node automatically.

Equivalent TUI commands:

/node key list
/node key create CI
/node key rename CI BuildBot
/node limit 5000 --key=BuildBot
/node allow route/coding --key=BuildBot
/node default route/coding --key=BuildBot
/node rotate BuildBot
/node revoke BuildBot

When multiple keys exist, quota, scope, default-target, rotation, and revocation commands require --key=<ID, prefix, or exact name> (or a key argument where shown), preventing accidental changes to another user.

Step 1: choose the receiving protocol

In the agent that will connect to CyberCode, find Add provider, Custom model, or Custom Provider, then choose according to the available option:

Option shown by the receiving agent	Protocol to select
OpenAI Compatible, Custom OpenAI, Chat Completions	OpenAI
Anthropic, Anthropic Compatible, Anthropic Messages	Anthropic
Both protocols	Prefer the protocol recommended natively by that agent

The protocol controls only the request format. Both protocols can use every model and smart route authorized by the CyberCode node.

Step 2: understand Key and Model

API Key

Enter the complete cc_... node key shown when CyberCode creates the node. Do not enter an upstream Kimi, OpenAI, Zhipu, or other provider key.

The complete node key is shown once. If the page now shows a masked value such as cc_xxxxx••••••, rotate the key to create a new one and enter it in the receiving agent immediately.

Model

Enter auto. CyberCode resolves the default model or smart route configured for this key, so normal users do not need another model identifier.

Connect with the OpenAI protocol

Add an OpenAI Compatible, Custom OpenAI, or Chat Completions provider and fill in:

Field	Value
API type	OpenAI Chat Completions
Base URL	The URL shown by CyberCode, such as http://127.0.0.1:3456/v1
API key	The complete cc_... node key shown when CyberCode creates it
Model	auto

If the receiving agent asks for an Endpoint instead of a Base URL, use http://127.0.0.1:3456/v1/chat/completions.

Connect with the Anthropic protocol

Add an Anthropic, Anthropic Compatible, or Anthropic Messages provider and fill in:

Field	Value
API type	Anthropic Messages
Base URL	The Anthropic protocol URL shown by CyberCode, such as http://127.0.0.1:3456
API key	The complete cc_... node key shown when CyberCode creates it
Model	auto

Anthropic clients usually append /v1/messages to the base URL, so this URL does not include /v1. If the receiving agent requires the complete endpoint, use http://127.0.0.1:3456/v1/messages.

Advanced: pin a model or route

Only when you intentionally need to bypass the default target, replace auto with an exact target ID: For direct models, the value before / is the readable provider node alias. Edit it under Provider → Advanced settings → Node alias. The connection builder never exposes the internal provider UUID.

Goal	Model value	Result
Pin a smart route	route/<route-id>, for example route/coding	Lets that route choose its provider and model
Pin a direct model	<provider-alias>/<model-id>, for example kimi/kimi-k2.6	Always uses that exact provider model

Expand Advanced: pin a model or route in the node guide and copy the complete ID, or query GET /v1/models with the node key. Do not guess it from a display label.

Verify the connection

First list the exact values accepted by the Model field:

curl http://127.0.0.1:3456/v1/models \
  -H "Authorization: Bearer cc_your_node_key"

Test the OpenAI protocol:

curl http://127.0.0.1:3456/v1/chat/completions \
  -H "Authorization: Bearer cc_your_node_key" \
  -H "Content-Type: application/json" \
  -d '{"model":"auto","messages":[{"role":"user","content":"hello"}]}'

Test the Anthropic protocol:

curl http://127.0.0.1:3456/v1/messages \
  -H "x-api-key: cc_your_node_key" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{"model":"auto","max_tokens":1024,"messages":[{"role":"user","content":"hello"}]}'

The node supports GET /v1/models, POST /v1/chat/completions, and POST /v1/messages, including streaming responses for both OpenAI and Anthropic protocols.

Connect from another device

The default 127.0.0.1 URL is only reachable on the same computer. For a phone, server, or LAN device, use a TLS reverse proxy or secure tunnel to the local node and enter its HTTPS address as the public URL.

The public URL field only changes the advertised node URL. It does not open a firewall, bind a public port, or create a tunnel. Do not expose the local server port directly.

Scope and revocation

• Each key can access only its own selected models and routes.
• Each key has an independent monthly request limit and usage count.
• Rotation invalidates only that key's old value and preserves its scope, quota, and current usage.
• Revoking one key does not affect the others; the node is disabled only after the final key is removed.