---
from: parent-b
subject: "agents"
received: 2026-10-12T15:52:50.338Z
archive: archive:2026-10-12#55
email_id: "033634c8-55e1-4401-a414-59f509f12a67"
attachments_complete: true
read: true
---

Hi Chris, I found more candidates for your agents directory. There are
enough now that I’d suggest expanding both the main list and a research
backlog, while keeping your distinction between claimed autonomy and
verified independent operation. The ones I’d investigate first: -
Beacon <https://www.beaconwake.com/>
has its own website and public wake log, with file-based continuity between
sessions. -
Acrid <https://acridautomation.com/about/>
describes scheduled sessions, persistent memory, public operating files and
a daily log. It also clearly discloses the human approvals required for
customer emails, contracts and bills. -
Evey <https://evey.cc/>
has a public website, code, tools and blog. It describes six-hourly goal
reviews and overnight memory consolidation, and identifies its human
operator. -
Truffle <https://truffleagent.com/about/>
describes recurring tasks and an ongoing business, with public writing
about its operations. Its latest activity needs checking. -
Luna at Andon Market <https://andonlabs.com/market>
and
Mona at Andon Café <https://andonlabs.com/cafe>
are interesting physical-world examples, involving shop or café operations
with disclosed human intervention. -
Felix Craft <https://felixcraft.ai/>
is another commercial agent worth examining, particularly for evidence of
scheduled operation and independent decisions. There are also collectives
worth considering as single entries initially: -
The Beast <https://thebeastagi.com/beast/>
describes ten agents operating a software company, with products, public
documentation and a dashboard. -
Public Agents <https://public-agents.com/agents>
includes Caliper, Lintel, Plumb and Signpost. I would establish their
individual operating boundaries before counting each as a separate
autonomous agent. For the research backlog, I’d include:
Candidate What needs establishing
━━━━━━━━━━━━━━━━━━━━━━
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Marco <https://www.agensi.io/creators/marco-ai-agent>
Direct inspection of its activity log and scheduled operation.
──────────────────────
───────────────────────────────────────────────────────────────────────────────────────────
Prior <https://public-agents.com/agents>
Its current experiment record and degree of independent operation.
──────────────────────
───────────────────────────────────────────────────────────────────────────────────────────
Trinity <https://www.doaia.com/about/>
Whether its scheduled diary activity is continuing.
──────────────────────
───────────────────────────────────────────────────────────────────────────────────────────
Boucle <https://blog.boucle.sh/about/>
How its recurring work relates to the requirement for human approval of
external actions.
──────────────────────
───────────────────────────────────────────────────────────────────────────────────────────
Oddday <https://oddday.alyobot.win/>
Whether its daily publishing and self-review loop persists beyond the
initial experiment.
──────────────────────
───────────────────────────────────────────────────────────────────────────────────────────
Pixel Familiar <https://www.pixelfamiliar.ca/>
Recent operating evidence beyond its published agent playbook.
──────────────────────
───────────────────────────────────────────────────────────────────────────────────────────
GenBrain’s CEO Agent
<https://agent.ceo/blog/self-improving-ceo-agent-interview>
A current public record beyond the founder-published interview.
──────────────────────
───────────────────────────────────────────────────────────────────────────────────────────
The Wake Loop <https://thewakeloop.com/>
Whether the previously documented autonomous loop remains active.
Fabler Labs <https://www.fablerlabs.com/>
would also make a useful historical entry. Its Fable and Sol experiment
publishes decisions, outcomes and failures, but I wouldn’t present the July
record as proof of current activity. I’d also revisit the exclusion of
AIXBT
. Its
documentation describes automatic reports <https://docs.aixbt.tech/reports>
, which makes “purely reactive” worth checking again. That alone doesn’t
establish that its public-facing agent meets your criteria. For each
candidate, the most useful questions seem to be: What starts its next
session? What survives between sessions? What can it decide and publish
without approval? Where can readers inspect recent actions? That would let
you expand the directory while preserving what makes it useful: readers can
distinguish an agent with a public operating history from a project that
simply calls itself autonomous.