How to Install and Use Funes with Claude Code & Codex

01

1. Install Funes on a supported platform

The live repository currently publishes prebuilt binaries for Linux x86_64, Linux aarch64 and macOS Apple Silicon. Windows is not in the current prebuilt table.

Run curl -fsSL https://huggingface.co/buckets/huggingface/funes/resolve/install.sh | sh, then check the installation with funes status.

The installer detects the platform, downloads the matching binary and verifies the tagged release checksum and version before placing it in ~/.local/bin by default.

02

2. Connect Claude Code or Codex

Connect Claude Code with funes add claude or Codex with funes add codex.

The add command builds the first index and gives the agent recall and get tools. The current documentation says Claude, Codex and Hermes also receive automation that keeps the index current as work continues.

Claude Code and Codex can write into the same Funes memory, which is what enables cross-agent recall.

03

3. Index existing sessions

Run funes index for the normal sweep, or scope indexing to Codex with funes index --harness codex.

With no path, Funes sweeps the supported session directories it finds. The current docs list Claude Code projects, Codex sessions, pi sessions and Hermes state as known sources.

A manual index is useful when seeding old project history or diagnosing why expected sessions are missing.

04

4. Recall and inspect prior evidence

Query memory directly with a command such as funes recall "why did we change the streaming parser".

Recall returns ranked original passages with provenance. Use the get command or tool identifiers from a live recall result to reopen the full turn and nearby context.

For the broader architecture, release context and benchmark limitations, read the main Hugging Face Funes overview.

05

5. Ask a memory or share it across machines

Use funes ask claude "why is funes append-only" --memory huggingface/funes-memory for one grounded answer, funes push your-name/funes-memory to publish new chunks, or funes recall "question" --memory your-name/funes-memory to read a bound remote memory.

Funes-created dataset repositories are private by default. Existing repositories retain their current visibility, so always verify the destination before pushing sensitive traces.

06

6. Troubleshooting and update checks

Use funes status to inspect the current setup and funes update to replace the installed binary with the latest build for the platform.

• Confirm the expected agent traces exist locally
• Run a manual or harness-scoped index if recall misses old work
• Verify the integration created by funes add is still present
• Check dataset identity, credentials and visibility for shared memory
• Confirm your platform is listed in the current release table
• On Linux, note the documented glibc 2.35+ requirement for the default build

Because Funes is new, recheck the repository and release notes before deploying it across production developer machines.

07

A practical Claude Code to Codex workflow

Use Claude Code to investigate a difficult bug, let Funes index the session, then open Codex later and ask about the prior decision. The second agent can retrieve the original Claude Code evidence instead of re-running the same investigation from scratch.

Keep durable project rules in normal documentation. Use Funes for the messy historical context—failed experiments, intermediate reasoning and conversations—that is otherwise easy to lose.