Tools
Tools
Connect your editor, terminal agent or SDK to Cognitivers.
Choose your tool for setup instructions, compatibility notes and a verification prompt. Each guide links to the official documentation used to check its configuration.
Find your integration
17 tools found
Terminal agents
OpenAI Codex CLIOpenAI's terminal agent and IDE extension. Custom providers are a first-class setting.Works directlyClaude CodeAnthropic's terminal agent. Speaks the Anthropic Messages API, so it needs a bridge.Needs a bridgeDeepSeek HarnessDeepSeek's agent harness. Providers are declared in one YAML settings file.Works directlyPiMinimal terminal coding harness. Same provider schema as DSH, in JSON.Works directlyOh my PiA fork of Pi with the editor wired in. Same schema, YAML, and model roles per task.Works directlyAiderPair programming in the terminal. Any OpenAI compatible endpoint, by environment or config file.Works directlyopencodeTerminal agent configured with a JSON file and the AI SDK.Works directlyGooseBlock's agent. Uses OPENAI_HOST rather than the OPENAI_BASE_URL name other tools use.Works directlyCrushCharm's terminal agent. The provider type matters: openai-compat, not openai.Works directly
Editors and IDE extensions
GitHub Copilot in VS CodeCustom Endpoint provider. Point it at the base URL and VS Code discovers the models itself.Works directlyZedThe most precise fit for our stack: it exposes the request-shape switches other editors hide.Works directlyContinue.devOpen source assistant for VS Code and JetBrains, configured with YAML.Works directlyClineVS Code agent extension with a generic OpenAI Compatible provider and a verify button.Works directlyRoo CodeVS Code agent extension, same provider shape as Cline, with stricter tool-call requirements.Works directlyCursorCannot use a third party endpoint. Its own documentation lists five providers and no base URL.Not supported
SDKs and proxies
OpenAI SDKs and TypeScript frameworksCode you write yourself. Every mainstream client takes a custom base URL, with one naming difference per language.Works directlyLiteLLMNot an agent. A local proxy that speaks OpenAI outward and Anthropic inward, which is what makes Claude Code usable.Works directly
What breaks, and why
Three things account for almost every failure, and none of them is the model:
- The base URL is missing
/v1. Most clients append the path themselves, so a base URL without it produces a request to/chat/completions, which does not exist. - The client sends the wrong request shape. Codex, opencode and the Vercel AI SDK can all be configured to send Responses-style requests by default. This API serves chat completions, and each of those tools has an explicit way to ask for it.
- The tool-call field is named differently. Some clients send
max_completion_tokenswhere the server expectsmax_tokens, and only Zed and Pi expose a switch for it. Where a tool does not, tell us: it is worth fixing on our side once rather than in every user's configuration.
Let the agent verify itself
For each of these tools there is a prompt you can paste into it, which makes the agent confirm the credential, the endpoint and streaming, and report the exact error if something fails. They are in Research prompts.