Skip to content

Memory and identity

An identity repo is a private GitHub repository that backs an agent’s durable identity: who it is, what it has learned, and how it works. Because identity lives in git, it is versioned, and it survives even a full teardown of the agent.

The repo holds the agent’s persona files, long-term memory in memory/, session state in state/ and .runtime/, and optional skills and configuration. Contrast this with the default: without an identity repo, an agent’s instructions are generated once at pod start, and learnings and session summaries are lost when the pod is replaced.

The template is runtime-neutral. AGENTS.md is the identity contract read directly by Codex, and CLAUDE.md is Claude Code’s entrypoint to the same contract, so shared memory, state, and skills work in either runtime. At boot, Kyber also renders the platform’s own manual into the repo’s .runtime/ directory: the platform explaining itself to the agent, matched to the version the agent is actually running on.

When you create an agent, you choose one of three modes:

  • Create new from template: Kyber scaffolds a private repo named <agent-name>-agent from a template, with no GitHub setup on your part beyond the one-time platform GitHub App registration.
  • Link existing repo: bring your own repo, fork another agent’s, or migrate an agent you set up by hand.
  • None: run the agent without a managed identity repo, exactly as before the feature existed.

The agent reads and writes its own identity repo with a short-lived token minted by the Kyber Platform GitHub App, scoped to just that repo. Nothing is baked into the agent’s git config, and an agent can only ever mint the token for its own repo. If the credential path breaks, git fails loudly instead of silently falling back to a broader token.

On every restart, the agent re-syncs its repo to the default branch, so it picks up what was merged while it was busy while keeping its own branches and uncommitted work. When you delete an agent, Kyber preserves the repo: memory and session history are valuable post-mortem, and deleting them is your call, not the platform’s.