Back to Adamant

Precise memory for Codex and Claude Code

Adamant gives CLI coding workflows a narrow RAG command that returns only the specific account, project, or repository context requested.

The simple workflow

Clone your own repository locally, start Adamant inside that project, pull your memory, index it, then let your coding tool call Adamant only when it needs targeted context.

git clone <your-repo-url>
cd <your-repo>
npx adamant-memory init
npx adamant-memory login
npx adamant-memory pull
npx adamant-memory index
npx adamant-memory context "what prior decisions affect this file?" --top 5 --max-tokens 1200

Why CLI retrieval

Broad context integrations can flood an agent with too much information. Adamant should work like a precise retrieval command: ask one question, search the RAG, return the top few cited memories, and keep the context window clean.

Three ways to use RAG

Pinecone-backed RAG

Reference your own Pinecone index and namespace for hosted vector search over memory and project context.

Local RAG stack

Build chunks, embeddings, retrieval, and graph context on your own machine for privacy-first or offline work.

Live Adamant connection

Connect to your Adamant account or local helper and retrieve memory on demand without copying secrets into your repo.

Useful commands

npx adamant-memory search "billing retries" --top 10
npx adamant-memory context "auth refactor constraints" --top 5 --max-tokens 1200
npx adamant-memory decisions --repo . --since 90d
npx adamant-memory sources "pinecone namespace migration"
npx adamant-memory status

What the coding tool can request

Relevant prior decisions before editing a file.
Architecture context from ChatGPT, Claude, notes, docs, and project history.
Past debugging attempts and known failure modes.
Product requirements, persona notes, and roadmap context.
Source provenance for retrieved memory.

Security model

Adamant is local-first. Provider keys are not stored in the repository. Account context, local RAG files, and Pinecone configuration stay under the user’s control. CLI commands can cap result count and token budget.

Who should use this

Adamant is for developers who use Codex, Claude Code, Cursor, or other AI coding workflows and want their tools to retrieve project history, architecture choices, personal preferences, and cross-model conversation context only when needed.