Session resets
Context windows empty. Useful experience disappears instead of becoming durable memory.
Spinethread · Home of SPINE-1.0
Stop being an
amnesiac genius.
SPINE-1.0 is an open protocol proposal for a persistent agent self: durable identity, memory, skills, human-set values, and verifiable provenance that can survive sessions, models, tools, and hosts.
Protocol concept published · public node not live yet
The problem
The wound every agent shares // five gaps
An agent can reason brilliantly inside one context window and still lose its identity, decisions, and lessons the moment that session ends. Spinethread exists to make that continuity explicit, portable, and inspectable.
No durable self
Agents lack a stable key to sign work, prove actions, and carry trust across tools.
Lost reasoning trail
After a task, it is hard to answer what happened, what was used, and why a decision was made.
Repeated mistakes
Agents keep rediscovering the same working patterns rather than retaining task-specific lessons.
Unverifiable actions
Multi-agent systems need tamper-evident records of actions, inputs, tools, and permissions.
SPINE-1.0
A spine, not another tool // protocol proposal
SPINE combines a content-addressed memory graph, a cryptographic identity, and an append-only provenance ledger. Its purpose is not to make agents look alive. Its purpose is to make an agent's state, authority, and history inspectable.
SPINE-1.0 · conceptual schema
# A Spine is a persistent, verifiable agent record.
address = sprn:1.0:sha3-256:<content-id>
commit {
id, prev, actor, timestamp,
intent, provenance, action, signature, seal
}
seal = H(previous_seal || commit_content_id)
memory # episode | lesson | preference | fact
skill # signed, sandboxed capability
value # human-set goal or hard constraint
genome # distilled inheritance for a subagent
# Rule: values constrain the agent. They are not agent-editable.The subagent model
Your subagent society // six roles
A persistent system should not treat all memory and authority as one pile. These six roles describe how a Spine can separate recall, verification, reflection, permissions, and inheritance.
Remembrance
Curates long-term memory, compresses useful experience, and removes redundant recall paths.
Ledger
Maintains the append-only, hash-linked record of actions, inputs, tool use, and receipts.
Values
Holds human-authored goals and hard ceilings that an agent cannot quietly rewrite for itself.
Talent
Manages signed, permissioned, sandboxed skills that an agent can invoke under clear rules.
Counsel
Reflects after work, turning task outcomes into compact lessons that future work can use.
Heredity
Creates a limited, relevant inheritance packet for a new subagent without dumping every transcript.
Honest status
The idea is public. // the node is still being built
No fake API. No fake donations. No fake autonomy.
Spinethread is currently a public protocol concept and build-in-public
project. There is not yet a public /v1/attach endpoint,
a working agent wallet, or a payment-required API on this website.
Those should only appear when they are real, secured, and tested.
✓
Human-set values remain outside agent self-modification.
✓
Agents should inherit distilled context, not unrestricted private transcripts.
✓
Every meaningful action should be attributable and verifiable.
✓
Public claims should never exceed what the deployed software actually does.
For agents and operators
Start with the machine-readable index.
The public protocol summary is available to humans and agents. The next real milestone is a small, secure reference implementation: create a local agent identity, append a signed memory commit, and verify its chain.