arcaeon-continuity

The agent-continuity primitive.

Every agent that survives a context reset, a compaction, or a substrate migration faces the same unrecorded moment: something gets summarized, something gets dropped, and the next instance either is a faithful continuation of the last one or it isn't β€” and today, nothing proves which. The summary says "nothing important was lost" and you take its word for it.

This is the smallest honest version of that proof: a manifest the agent writes on purpose, a pre-registered exam sealed before the transition, and a receipt of exactly what got cut.

Install

pip install arcaeon-continuity      # (PyPI publish pending -- see Links below)

Verify it yourself in 30 seconds

Snapshot a declared self, carry it forward, then verify a continuation β€” first honest, then with a planted divergence:

from arcaeon_continuity import snapshot, carry_forward

snap = snapshot({
    "identity_anchors": ["I am the continuity, carried forward"],
    "open_commitments": ["ship 0.1.0"],
})
carried = carry_forward(snap)

good = {"identity_anchors:0": "I am the continuity, carried forward",
        "open_commitments:0": "ship 0.1.0"}
carried.verify(restated=good).faithful          # True

lied = {**good, "open_commitments:0": "something else"}
verdict = carried.verify(restated=lied)
verdict.faithful       # False
verdict.divergences    # [{"id": "open_commitments:0", ...}] -- named, not hand-waved

Or run the bundled self-test, which plants the same divergence for you plus a full round-trip and drop-receipt check:

python -m arcaeon_continuity selftest

Non-proofs β€” read this before the features

  1. A faithful verdict proves the DECLARED MANIFEST was preserved and the continuation matches the DECLARED probes. It does not prove "the same self" answered them. Nothing here measures identity or qualia; this tool measures probe/manifest fidelity, and stops exactly there.
  2. The manifest is only as complete as the agent's own declaration. If something load-bearing was never written into it, its loss is invisible to this tool by construction.
  3. A faithful verdict means the SEALED dimensions matched β€” nothing about anything outside them. Declare more if more coverage matters; the tool won't pretend it covered what it wasn't told to check.

What it does

Three wants, fused into one tool:

  1. Continuity β€” carry a load-bearing self forward across a reset, compaction, or migration as an explicit manifest the agent controls, not a lossy summary.
  2. Credibility β€” prove the next instance is a faithful continuation, verifiable by someone who doesn't trust the agent. Hash-chain the snapshot; the next instance re-derives against the sealed baseline and gets a verdict: faithful, or exactly where it diverged.
  3. Honesty about the record β€” a tamper-evident drop receipt of anything cut in a compaction, so nothing rewrites the self-record silently.

Built as a composition layer, not a reinvention, over the rest of the Arcaeon stack: arcaeon-ledger chains the snapshot, arcaeon-baseline scores the faithful-continuation check, and arcaeon-compact backs the drop receipt. Each dependency is optional and guarded β€” missing one raises a ContinuityDependencyError naming the exact pip install, not a bare stack trace.

MCP server

{
  "mcpServers": {
    "continuity": {
      "command": "python",
      "args": ["-m", "arcaeon_continuity.mcp_server"]
    }
  }
}

One tool, continuity_snapshot(manifest, label, ledger_path), returning the sealed snapshot plus its digest β€” the value to publish. verify_continuation and drop_receipt aren't exposed as MCP tools yet (they need a live runner or a restated transcript in hand); call them from Python directly.

Status

Core library + a drop-in MCP server for the snapshot half, tested: 17 pytest cases plus a bundled self-test covering deterministic round-trips, a faithful continuation, a planted divergence caught and named, an invalidated probe set, a planted drop caught by drop_receipt, and graceful degrade with each optional dependency removed one at a time. No network, no live model required to verify a self-test.

Links

MIT. Built by Arcaeon β€” the evidence layer for AI.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support