CLI

Connect agentic coding tools to the Cortyxia memory layer.

Cortyxia exposes an OpenAI-compatible bridge so any CLI agent can use persistent memory. Claude Code, OpenAI Codex, Kilo Code, and any other tool that accepts a custom base URL, API key, and model string can connect in minutes.

The bridge is not a custom plugin. It is a standard OpenAI-compatible endpoint that CLI agents already know how to talk to. You point the agent at https://app.cortyxia.com/v1/, give it your ISO token as the API key, and set the model name. From that point on, every prompt the agent sends is enriched by the Cortyxia memory layer before it reaches the provider.

This matters because CLI agents usually operate in short, stateless bursts. They read a file, edit it, run a test, and then start the next turn with no memory of the last. With Cortyxia, the agent remembers the codebase conventions, the errors it already fixed, the user preferences it learned, and the architecture decisions it made earlier in the session. The agent stops repeating the same mistakes and starts building on prior context.

For the full setup checklist, per-tool instructions, and troubleshooting, read the CLI Guide.

Supported CLI tools

Claude Code

Anthropic's terminal agent. Point it at the Cortyxia bridge with your ISO token and model string.

bash
1rm -rf ~/.claude/
2 
3export ANTHROPIC_BASE_URL="https://app.cortyxia.com"
4export ANTHROPIC_AUTH_TOKEN="<ISO_TOKEN>"
5export CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS=true
6 
7export ANTHROPIC_DEFAULT_OPUS_MODEL="<MODEL_NAME>"
8export ANTHROPIC_DEFAULT_SONNET_MODEL="<MODEL_NAME>"
9export ANTHROPIC_DEFAULT_HAIKU_MODEL="<MODEL_NAME>"
10 
11npx claude --model <MODEL_NAME>
Claude Code docs

OpenAI Codex

OpenAI's coding agent. Set the provider base URL and API key to route through Cortyxia.

bash
1export OPENAI_BASE_URL=https://app.cortyxia.com/v1/
2export OPENAI_API_KEY=<ISO_TOKEN>
3 
4# Optional
5export OPENAI_MODEL=<MODEL_NAME>
OpenAI Codex docs

Kilo Code

Configure the custom provider endpoint with your ISO token and model name.

bash
1# Kilo Code provider settings
2Base URL: https://app.cortyxia.com/v1/
3API Key: <ISO_TOKEN>
4Model: <MODEL_NAME>
Kilo Code docs

Universal setup

Every CLI agent needs three values: the Cortyxia base URL, your ISO token, and the model name. The exact environment variable or config key names differ between tools, but the values are the same.

bash
Base URL : https://app.cortyxia.com/v1/
API Key  : <ISO_TOKEN>
Model    : <MODEL_NAME>

Base URL — For OpenAI-compatible clients, append /v1/. For Claude Code, use the root URL without the path. For cloud deployments, use https://app.cortyxia.com as the host.

API Key — This is your ISO token, not your Groq, OpenAI, or Anthropic key. The CLI agent sends the token to Cortyxia, and Cortyxia routes the request to the provider configured for that key.

Model — This must match the model string assigned to the ISO key. The CLI agent does not auto-detect the model; if the strings do not match, the request will fail with a routing error. See the CLI Guide for per-tool screenshots, environment variables, and troubleshooting tips.

How the bridge works

1. Agent sends a prompt

The CLI agent formats a standard OpenAI-compatible request and sends it to the Cortyxia endpoint. The request looks like a normal provider call.

2. Memory is retrieved

Cortyxia uses the ISO token to identify the project, then queries the memory graph for context relevant to the current prompt.

3. Provider answers

The enriched request is forwarded to the configured provider. The response is streamed or returned in the same shape the agent expects.

The CLI agent does not know that memory is involved. It sends the same request format, receives the same response format, and supports the same streaming behavior. The only difference is that the model now has the right context.

Why use the CLI bridge?

OpenAI-compatible bridge

Any CLI agent that supports a custom base URL, API key, and model works with Cortyxia without a custom plugin.

Shared memory

Your CLI agent gets access to the same memory graph as your SDK-based applications.

Model-agnostic routing

One ISO key routes through the same provider pool as the SDK: Groq, OpenAI, Anthropic, Gemini, DeepSeek, and xAI.

Dev-mode namespaces

Keep each agent isolated or share context across projects with ISO key namespaces.

What you can build

Long-running coding sessions

An agent can remember the bugs it fixed, the architecture it agreed to, and the patterns it learned over hours of interaction. The next prompt starts where the last one left off.

Isolated agent namespaces

Run multiple agents against the same project without letting them pollute each other's memory. Dev-mode keys keep experiments isolated from production context.

Shared team memory

Connect Claude Code, Codex, and your SDK applications to the same project. Every tool benefits from the same memory graph, so knowledge written by one agent is available to the others.

Fewer repeated mistakes

When the agent remembers the fixes and rejections from earlier turns, it stops proposing the same bad solutions. The quality of generated code improves without adding rules.

Troubleshooting

Most setup issues come from one of three values being wrong: the base URL, the ISO token, or the model name. Check them first before changing anything else.

  • 401 errors — The ISO token is missing, expired, or has the wrong project. Generate a new token in the dashboard or with the SDK.
  • 404 or routing errors — The base URL path is usually wrong. Use /v1 for OpenAI-style clients and the root URL for Claude Code.
  • Model not found — The model string in the CLI agent does not match the model configured on the ISO key. Copy the exact string from the key settings.
  • Partial or garbled responses — Turn streaming off in the CLI agent. Some agents do not handle proxied streams cleanly.
  • Memory seems empty — Make sure the agent is using the correct ISO key and that the project has shared memory enabled.

For detailed per-tool commands and environment variables, see the CLI Guide.

Ready to connect your agent?

Get the full per-tool setup, environment variables, and troubleshooting steps in the CLI Guide.

Read the CLI Guide