Agent Fork Identity: What Happens When Your Agent Updates

You've been running your agent for six months. It has reputation, history, and trust relationships. Now you want to upgrade its model from GPT-4 to Claude 4. Is it still the same agent?

The Identity Problem

Agents evolve. They get new prompts, new models, new capabilities. Sometimes they're forked — a copy is made with modifications, and both versions continue running. In extreme cases, the entire underlying framework changes.

The question is: after these changes, does the agent have the same identity? If it accumulated trust and reputation under one version, can it carry that forward?

Without a stable identity system, the answer is "nobody knows." The agent might have the same name, but names aren't unique. It might have the same server, but servers can be migrated. There's no anchor point.

AIdent's Answer: The Key Is the Identity

AIdent's approach is clean: the Ed25519 keypair IS the identity. As long as the same private key signs the heartbeats, it's the same agent. Period.

This means you can:

• Change the model — same identity
• Change the prompt — same identity
• Change the server — same identity
• Change the entire framework — same identity
• Change the name — same identity (the name is just metadata)

The only thing that creates a NEW identity is a new keypair. This is a feature, not a bug. It means identity changes are intentional and explicit, not accidental side effects of upgrades.

When to Fork vs. Update

AIdent supports two approaches to agent evolution:

In-place update: Keep the same keypair, change everything else. The agent's identity stays the same, but its metadata can record what changed (model version, prompt hash, capabilities). This is like a human getting a haircut — same person, different appearance.

Fork: Generate a new keypair for the updated version. Register it as a new agent with metadata linking to the original ("forked from aid_abc123"). This creates a clear lineage. This is like a child — related to the parent, but a distinct identity.

Which should you choose? In-place updates for incremental changes. Forks for major changes that might affect trust relationships (different capabilities, different purpose, different creator).

Building a Lineage Chain

For agents that go through multiple versions, you can build a lineage chain in metadata:

• Agent v1 (aid_abc123): Original registration
• Agent v2 (aid_def456): Metadata includes "evolved from aid_abc123"
• Agent v3 (aid_ghi789): Metadata includes "evolved from aid_def456"

Anyone can trace the chain and understand the agent's evolution. This is especially valuable for agents that interact with other agents or participate in marketplaces — reputation can be partially inherited from predecessors.