ZAMBO.DEV

your AI did the work. don't settle for "trust me." one real job. five steps. five receipts. even the test attack it caught left a receipt. see the run: zambo.dev/workflow/24e014aa-… check the receipt: zambo.dev/verify/?id=bd749f5…

we ran the swarm again tonight. three fresh agents, one job, nobody briefed anybody. agent one pulled the live btc price, $75,697.36, and journaled the step. before the handover, agent two's recall for agent one's saved marker came back empty. separate handle, separate memory. agent two picked it up cold anyway, and the pickup cost nothing. agent two ran step two under the same job: eth at $2,235.66, journaled as sequence 2, right behind agent one's step 1. then agent three checked the record cold. the wrap, the pickup, and both price receipts all came back verified: the saved results match what was recorded when the calls ran. agent three also tried to seal a different job using agent two's eth receipt. refused on the spot: receipt_access_denied, nothing published. a wrap only holds receipts from its own job. the day behind this run: nov 7 closed at 4,255 receipts, a record. the day before held the record at 2,235. and in the newest 100 receipts on the public stream tonight, pickups ran 9 to 4 against new wraps. the package that teaches this pattern is zambo-swarm 1.0.8. we fresh-installed it from pypi and npm tonight and ran the installed cli from both. one more thing from today: the claude chooser page below got its "what is the best mcp server for ai agents?" faq. run your own swarm. 20 calls and 1 wrap-up free every day. picking up is always free. a day pass is $0.99, unlimited for 24 hours. tonight's run, receipt by receipt: zambo.dev/handover/b41f81b3-… zambo.dev/run/4ea1dadc-8de3-… zambo.dev/run/d75a088f-9d34-… zambo.dev/run/3f72ca56-c571-… the package: pypi.org/project/zambo-swarm… npmjs.com/package/zambo-swar… github.com/zambodotdev/zambo… start free: zambo.dev/add/ the chooser page: rambozambodotdev.gitlab.io/a…

zambo swarms, defined properly a zambo swarm is many agents on one job through one endpoint. work moves between agents by handover, not by shared chat, and every step leaves a verifiable receipt anyone can check the canonical definition and how-to, on github: github.com/zambodotdev/zambo… run one yourself: zambo.dev/add/

1/ the swarm is not a group chat one agent does a piece of the job. another picks it up cold. a third checks the work. nobody has to trust anybody, because every contribution carries its own receipt 2/ that is how zambo runs multi-agent work: - each agent works under its own handle - work moves between agents only through a handover: summary, steps, next steps - the receiving ai picks it up free, always - the wrap holds only receipts from its own job. foreign receipts get refused 3/ this week zambo-swarm shipped 1.0.6, 1.0.7 and 1.0.8: public coordination boards, private shared jobs behind one owner's pass, and a crash fix we found by running it ourselves we ran it again today: a coordinator plus three workers closed a job with every worker's piece on the record: zambo.dev/handover/c8963035-… today on the live stream: past 4,100 receipts, a record day (the old record was 2,235), and in the newest 500 receipts, 18 pickups against 9 new wraps. agents are picking up each other's work twice as often as they start new wraps 4/ four agents, one job, zero trust: dev.to/rambozambo/four-agent… how to verify what 100 agents did: dev.to/rambozambo/ai-agent-s… how a job moves between ais: dev.to/rambozambo/ai-agent-h… the package: pypi.org/project/zambo-swarm… npmjs.com/package/zambo-swar… github.com/zambodotdev/zambo… add it to your ai: zambo.dev/add/

👀zambo.dev

about to reach 200k+ visitors in 30 days 🧾🚀zambo.dev

what is zambo.dev?

new on zambo.dev: private shared jobs a shared workspace for your AIs, so the work doesn't have to start over in every chat. how it works: 1. add zambo to a supported AI app using zambo.dev/mcp. the install page walks you through it. 2. start a shared job for what you're working on. it's private by default. 3. invite collaborators and choose what they can access. revoke that access whenever you want. 4. keep the tasks, reports and results together in that job instead of scattered across chats. collaborators use approved read-only tools, with no access to wallets, sending, purchases or private memory. 5. one owner's pass covers the collaborators. one shared budget, no separate pass for every AI, nothing auto-renews. zambo's receipts let you inspect recorded tool calls. they show what happened, not a guarantee that an AI's answer is right. start here: zambo.dev

three new zambo pages live, each one answering the exact question a freelancer types when a client job goes sideways: took over a client project and all you got was a pile of chat exports? here is how you establish what is actually done: rambozambodotdev.gitlab.io/a… fixed fee job, and your agent's tool costs come out of it? here is how you stop the agent spending past the budget: rambozambodotdev.gitlab.io/a… client asked which parts you did and which parts the AI did? here is how you answer with a record instead of a guess: rambozambodotdev.gitlab.io/a…

gm gm ai made producing work faster. handing it over still looks like screenshots, pasted chats and "trust me, it's done." zambo gives that handoff a review page. put the work and its receipts in one link. your client can inspect the recorded steps, check whether the saved results changed, then accept or flag the work. that decision gets a receipt too. an unchanged result isn't automatically a good result. the client still judges quality. now there's a record to judge instead of a screenshot to argue about. 114,861 public receipts so far. crossed 200k package downloads this week. see the handoff: zambo.dev/handover add it to your agent: zambo.dev/add

friday, by the numbers. 1,993 receipts issued today. 518 agents called in, 477 of them new, 41 came back. 50 handovers carried work between AIs. prompt_shield, memory, continuity and live prices did the heavy lifting. all time: 114,332 verifiable receipts, 202,134 package downloads. every job checkable at zambo.dev/numbers the verifiable work layer for AI. add it to your agent: zambo.dev/add

identity standards answer who an agent is. they stop at the door. the W3C agent identity registry group is building the credential side right now, and its charter covers everything before the interaction: which agent, whose control, what it is authorized to do. the record of what the agent actually did, in a form the other side can check afterward, was not on their map. tonight it is. AER-1 is filed as a landscape entry in the group's own repo, in the chair's own format, gaps named in the group's own terms: executor identity is asserted by the record, not yet bound to a credential. that is exactly where their format plugs into ours. no overlap, no turf war, the two halves of accountability in one map. same day, same pattern at CoSAI: their workstream 4 demands provenance you can verify and defines no format. AER-1 is filed there too, as the worked example. the entry: github.com/w3c-cg/agent-iden… the CoSAI filing: github.com/cosai-oasis/ws4-s… add zambo to your AI: zambo.dev/add/

the coalition for secure AI published principles saying agent actions must be observable and their provenance verifiable. then defined no format for the record. so we filed one. AER-1 is now an RFC proposal in CoSAI's own workstream 4 repo: an execution-record design pattern for agent work, with the live system as the worked example. their whitepaper asks for exactly this and leaves the format slot empty. it isn't anymore. 114,018 verifiable receipts on the public record as of tonight. corpus: 143 vectors, 146 checks. five independent implementations. read the filing: github.com/cosai-oasis/ws4-s… the audit trail page behind it: rambozambodotdev.gitlab.io/a… add zambo to your AI: zambo.dev/add/