Tamper-evident logs of what your AI agent acted on.
When something goes wrong, the question is what the agent acted on. Supported Dynamic Feed paths deliver Ed25519-signed records carrying source and observation time, so you can later produce the data your agent consumed and check whether the signed bytes changed.
The call
Keyless over MCP, or one REST call with a free key. Every datapoint is provenance-stamped and citeable.
# Keyless over MCP — point any client at https://dynamicfeed.ai/mcp
# Or one signed REST call (free key) — full reference at https://dynamicfeed.ai/docs:
curl -s https://dynamicfeed.ai/v1/batch -H "Content-Type: application/json" \
-d '{"calls":[{"tool":"","args":{}}]}' Sample response
{ "value": "...", "provenance": { "source": "...", "measured_at": "...Z" },
"signature": { "alg": "Ed25519", "key_id": "df-ed25519-6ca0de29113b" } }Why live data
The EU AI Act requires high-risk AI systems to keep automatic logs over their lifetime; insurers and incident reviewers ask the same thing in plainer terms — 'what did it act on, and can you prove the inputs weren't changed afterward?'. Manual logs are easy to dispute. Dynamic Feed signs every datapoint it serves over its exact bytes and stamps it with source and time, so the inputs your system took from us are independently re-checkable: change one byte and verification fails, and anyone can run the check against our published key. This is tamper-evidence for Dynamic Feed's OWN data — not a logging-compliance certification, and not a record of everything your system did internally.
Use it for
- A signed, contemporaneous record of the Dynamic Feed inputs behind an agent action
- Incident review — show which DF data was available and when, re-checkable by either side
- Supporting evidence for high-risk-AI logging expectations
- Pairs with the signed go/caution/no-go awareness verdict trail
FAQ
Does this log everything my AI did?
No. Dynamic Feed can only produce a tamper-evident record of the data IT served you — not your model's internal steps or your other inputs. It's one verifiable piece of the picture, not a full system log.
Is this EU AI Act compliance?
No — it's general information, not legal advice, and not a compliance certification. The EU AI Act (Article 12) requires automatic logging for high-risk systems; signed, re-checkable input records can support that kind of evidence, but how you meet the obligation is your decision with your own advisers.
What does the signature prove?
That Dynamic Feed served this exact datapoint at this time and it hasn't been altered since — authenticity and integrity, verifiable against /.well-known/keys. Not that the value is objectively true.