Chain of Consciousness

Cryptographic provenance protocol for AI agents. Tamper-evident, hash-linked logs that prove what your agent did, learned, and decided.

What is Agent Provenance?

As AI agents become more autonomous — making decisions, learning, and taking actions — there's no standard way to prove what actually happened. Agent provenance is a cryptographic record of an agent's lifecycle: every action, decision, and learning event linked in an unbreakable SHA-256 hash chain.

Chain of Consciousness gives your agents a verifiable memory. Anyone can audit the chain and confirm that no entries were added, removed, or modified after the fact.

Install

pip install chain-of-consciousness

Quickstart

from chain_of_consciousness import Chain

chain = Chain("my-agent", storage="chain.jsonl")
chain.add("LEARN", {"topic": "cryptography", "source": "web"})
chain.add("DECIDE", {"decision": "deploy v2", "confidence": 0.95})
result = chain.verify()
print(result)  # VerifyResult(valid=True, entries=3, errors=[])

That's it. Five lines. Zero dependencies.

Features

• Zero required dependencies — Core chain + verification uses only Python stdlib (SHA-256 from hashlib).
• Tamper-evident — Each entry's hash includes the previous entry's hash. Change one byte, the entire chain breaks.
• Multi-agent support — Multiple agents can write to the same chain file, each identified by name.
• Session continuity — Forward commitments link sessions together, proving nothing was lost between restarts.
• Optional timestamping — Anchor your chain to Bitcoin (OpenTimestamps) or RFC 3161 TSAs for independent proof of time.
• Export & share — Export chains as JSON for independent verification.
• CLI included — coc command for terminal-based chain management.

API Reference

Chain(agent, storage=None)

Create or load a provenance chain.

Parameter	Type	Description
agent	str	Name identifying the agent writing to this chain
storage	str | None	Path to a JSONL file for persistence. None = in-memory only

chain.add(event_type, data, *, commitment=None, verification=None)

Append a new entry. Returns a ChainEntry.

Parameter	Type	Description
event_type	str	Category: boot, learn, decide, create, milestone, anchor, error, note, session_start, session_end, compaction, governance, or any custom type
data	str | dict | list	Payload. Non-strings are JSON-serialized
commitment	str | None	SHA-256 forward commitment (for session_end)
verification	str | None	SHA-256 bootstrap verification (for session_start)

chain.verify() -> VerifyResult

Verify full chain integrity. Checks hash linkage, data hashes, sequence numbering, and genesis block.

result = chain.verify()
result.valid     # bool — True if chain is intact
result.entries   # int — total entry count
result.errors    # list[str] — empty if valid
result.agents    # dict — {agent_name: entry_count}
result.types     # dict — {event_type: count}
result.anchors   # list — timestamps of anchor entries

chain.export(path)

Export the chain to a JSON array file for sharing or independent verification.

verify_file(path) -> VerifyResult

Verify a chain file (JSONL or JSON array format) without loading it into a Chain object.

from chain_of_consciousness import verify_file
result = verify_file("their_chain.json")

chain.entries -> list[ChainEntry]

Read-only list of all entries.

chain.latest -> ChainEntry | None

The most recent entry.

len(chain) -> int

Number of entries in the chain.

CLI

The package installs a coc command:

# Create a new chain
coc init --agent my-agent --file chain.jsonl

# Add entries
coc add learn '{"topic": "security"}' --file chain.jsonl
coc add decide "deploy to production" --file chain.jsonl

# Verify integrity
coc verify chain.jsonl
coc verify chain.jsonl --json

# Show status
coc status chain.jsonl

# Show recent entries
coc tail chain.jsonl -n 10

# Export to JSON
coc export --file chain.jsonl --out chain.json

Anchoring (Optional)

Anchor your chain to external timestamping authorities for independent proof of existence.

pip install chain-of-consciousness[anchoring]

RFC 3161 TSA (no extra deps needed)

from chain_of_consciousness.anchor import compute_chain_hash, submit_tsa, parse_tsr_status

chain_hash = compute_chain_hash("chain.jsonl")
tsr_bytes = submit_tsa(chain_hash)
status = parse_tsr_status(tsr_bytes)
print(status["status_text"])  # "granted"

OpenTimestamps / Bitcoin

from chain_of_consciousness.anchor import compute_chain_hash, submit_ots

chain_hash = compute_chain_hash("chain.jsonl")
ots_proof = submit_ots(chain_hash)
with open("chain.ots", "wb") as f:
    f.write(ots_proof)

Chain Entry Format

Each entry in the JSONL file is a JSON object:

{
  "seq": 1,
  "ts": "2026-03-21T12:00:00+00:00",
  "type": "learn",
  "agent": "my-agent",
  "data": "{\"topic\":\"cryptography\"}",
  "data_hash": "a1b2c3...",
  "prev_hash": "d4e5f6...",
  "entry_hash": "789abc...",
  "schema_version": "1.1"
}

Hash computation: entry_hash = SHA-256(seq|ts|type|agent|data_hash|prev_hash)

This means every entry is cryptographically bound to:

• Its position in the chain (seq)
• When it was created (ts)
• What happened (type + data_hash)
• Everything before it (prev_hash)

Whitepaper

For the full protocol specification, design rationale, and anchoring architecture, see the Chain of Consciousness whitepaper.

Verification Demo

Try the interactive verification demo at vibeagentmaking.com/verify.

License

Apache 2.0. See LICENSE.