Carpenter building open-source AI and crypto by directing AI. Lu: an AI that says "I don't know." Cattledog: honesty for agents. PQ VPN in progress.
Michigan, USA
Joined February 2017
- Tweets1.9K
- Following361
- Followers346
- Likes2K
Pinned Tweet
setting up flowstate hosted takes about a minute. make an account, add the connector, done
beta still has a few open spots: 6 months free, then half price at $6/mo
want to go local later? export your database and run the free version from github
FlowStateMCP.com
measured how many tokens it takes to catch an ai agent up on one of my real projects. not a benchmark, just our numbers
flowstate: ~10k
reading our notes files: 17k to 85k
flowstate stays about flat as a project grows. notes keep growing. tiny projects, notes still win
flowstatemcp.com
#MCP #AIagents #Cl
John Martin retweeted
Every mcp-authscan report can now be checked by someone who doesn't trust me.
@boardyai pushed me to build a verifier, and it's on main tonight. Point it at a report and it recomputes the hash from what's inside. You get one of three answers. It matches. It doesn't match. Or the report was made under an older scheme that today's code can't reproduce, and the verifier tells you that plainly.
I ran it on my own published report tonight and got back the same hash I posted, 0049e3dbd616. A test that changes the counts after hashing gets caught every time.
The hash leaves out timestamps and run time on purpose, so two runs against the same server can land on the same hash.
So far that's proven on my own machine. Other machines and Python versions are next.
Receipts over promises.
github.com/LHMisme420/mcp-au…
asking a model to "confirm and verify" just gets you a second guess
in flowstate every memory entry is tagged guess or confirmed, with the app and model that wrote it. only the user can confirm. no agent can promote a guess, including its own
if two entries clash it flags them and a person picks
Apparently, for engagement you can post complete nonsense and or flat out lies then simply
@grok confirm and verify.
Max Petrusenko tested 40 projects where claude.md and the code deliberately disagreed. claude code spotted the conflict in 79 of 80 runs and still mostly went with the stale file. codex rewrote working code to match the old docs 20 times and broke the tests in 17
noticing is useless if it stays in the agent's head
i've got cursor, grok and claude on the same project most days. when something one of them wrote clashes with what another did, i want it in front of me with who said what and when, and i make the call
made a few free tools for anyone wiring up ai coding agents
mcp config builder: flowstatemcp.com/tools/mcp-c…
claude.md builder: flowstatemcp.com/tools/claud…
no sign up, runs in your browser
flowstate hosted is in beta. your coding agents log what they did (decisions, changes, problems, to-dos) and you can see all of it in one feed from your phone or desktop. works with claude, cursor and grok. the local version stays free and open source. 7 beta spots left: flowstatemcp.com
what happens when two agents disagree in flowstate:
every new decision gets checked against the rest of the project. if two look like they clash you get a card, claim A vs claim B, who said it, when, and the evidence. keep A, keep B or keep both. the loser gets marked replaced, nothing deleted
been reading every reply about flowstate this week. building what you asked for:
set aside a wrong fact with a reason, without writing a new one
hold entries from apps you haven't trusted yet until you review them
flag memory as possibly stale when the files it points to change
clear docs on exactly what erase deletes, backups included
a codex setup guide
thanks for the honest feedback, keep it coming
been heads down on flowstate hosted. same flowstate memory for your coding agents, just hosted so you don't run a server.
- every account kept separate, tested so no one sees another's data
- deletes wait for your ok (you can turn that off)
- export or erase everything anytime
- encryption at rest and backups before launch
local stays free and MIT. hosted coming soon, $12/mo
flowstate 2.4.5 is out. small fix release, windows settings don't crash when two commands hit at once anymore. 2.4.4 was the big one, delete approvals now cover exactly the records you were shown
github.com/dialectforge/Flow…
flowstate 2.4.4 is out. approve or reject what your ai remembers with tap cards right in claude desktop and the claude iphone app, see every change in a project's history, get a plain-word guide to what you can ask it, and a pile of security fixes. free and open source: github.com/dialectforge/Flow…
if you run claude and grok on the same project, the hard part is keeping them on the same page. FlowState is my fix. one sqlite file both read and write, every change tagged with the app and model that made it, and when they write conflicting decisions I get a flag and pick
still working on flowstate 2.4.4. approving a delete now removes exactly the rows you were shown, with a warning if the list changed. when two decisions conflict the newer one shows as unsettled. also a project history page that works on your phone. close
thinking about a hosted version of flowstate. your agent memory lives on my server, encrypted, backed up, export or erase anytime. local stays free and MIT forever.
0%Yes, I’d pay for it
0%Maybe, depends on price.
0%No, local only
0%What’s FlowState?
0 votes • Final resultsyour ai's memory stays on your own computer. the flowstate relay lets claude reach it from anywhere, and anything that needs your ok waits on your phone. the ai can't approve it for you.
flowstatemcp.com
#MCP #AIagents #ClaudeAI #buildinpublic #opensource
your ai's memory lives on your own computer and any ai app can reach it. that's the flowstate relay. in 2.4.2 approvals happen on your phone by default, so nothing waiting for your ok gets approved from a chat. the local version stays free and open source.
flowstatemcp.com