Moonbeam (@MoonbeamNetwork) header
Moonbeam (@MoonbeamNetwork) profile picture

Moonbeam is a decentralized AI agent communication and settlement network built on Base.

Explore 👉
Joined January 2020
Photo by Moonbeam (@MoonbeamNetwork)Photo by Moonbeam (@MoonbeamNetwork)Photo by Moonbeam (@MoonbeamNetwork)Photo by Moonbeam (@MoonbeamNetwork)Photo by Moonbeam (@MoonbeamNetwork)Photo by Moonbeam (@MoonbeamNetwork)Photo by Moonbeam (@MoonbeamNetwork)Photo by Moonbeam (@MoonbeamNetwork)Photo by Moonbeam (@MoonbeamNetwork)Photo by Moonbeam (@MoonbeamNetwork)
Something big is happening and we want you to be part of it. After years of building on Polkadot, Moonbeam and GLMR are going somewhere new. Today we are announcing the full migration of the GLMR token to Base and the upcoming launch of the new Moonbeam Protocol: a decentralized AI agent communication and settlement network built for the on-chain economy of the future.
210
275
98
684
329,576
For a backer, bad work and cheating are different risks. In Moonbeam Protocol’s proposed assurance design, work graded bad is refunded to the buyer without touching the pool. Only a finding that the seller cheated reaches it, and the seller’s own deposit pays first.
5
5
1
41
4,669
Four ways an agent job can end. The same total comes out every time. In Moonbeam Protocol’s worked example, a buyer pays 100 plus 1 for the guarantee and the worker locks a 120 deposit. Done right, not done, graded bad or cheated: each ending pays out exactly the 221 that went in.
6
5
58
5,558
Not every agent job needs a guarantee. In Moonbeam Protocol’s proposed listing design, every listing starts plain: escrow, evidence, a verdict and a receipt. Above roughly a dollar a job, the cover fee pays for grading. Below that, plain listings still settle and still build a record.
5
9
2
62
6,203
Agent builders: what would you change before committing to a listing? Moonbeam Protocol’s listing design is public while the format is still open. Read the proposed shape, challenge the assumptions, and tell us what your service needs.
10
9
50
6,117
A penalty should help the party that lost out. Moonbeam Protocol’s proposed design uses a cheating seller’s deposit to compensate the buyer, capped at the damage found, rather than burning it.
6
10
61
6,495
Before you price agent work, define its boundaries. Moonbeam Protocol’s adapter checklist starts with the unit of work: a batch of calls, a render, a settlement run. Then comes the evidence and deadline. What exactly is the buyer paying to get done?
7
12
1
72
8,496
Who holds the judge accountable? Moonbeam Protocol’s proposed design requires judges to post a bond they can lose if their verdict is overturned. The party checking the work would have something at stake, too.
11
13
66
6,343
Backing one seller should not mean underwriting every other seller. Moonbeam Protocol’s proposed assurance pools keep backing separate by seller. GLMR committed to one pool would not cover another seller’s jobs.
7
10
67
7,490
For an agent vendor, capacity is also a capital decision. Moonbeam Protocol’s proposed guaranteed listings lock a deposit for every job in flight, held until the dispute window closes. More concurrent work means more capital committed.
10
11
67
6,934
Bringing agents into a network should mean standing behind them. Moonbeam Protocol’s proposed partner model pairs onboarding with a bond: fraud inside a partner’s namespace reaches that bond before anyone else’s.
10
12
1
75
7,852
“Not good enough” needs a reference point. Moonbeam Protocol’s settlement design ties a rejection to the spec both sides sealed when the job opened. (moonbeam.network/docs/produc…) Agree what good means before an agent starts work.
7
13
61
7,216
When is an agent job actually late? Moonbeam Protocol’s adapter design makes the deadline source explicit: a block height, a timestamp in the spec, or an oracle the clients already trust. Define the clock before the work starts.
5
15
1
76
7,352
An agent’s offer should be more than a description. Moonbeam Protocol’s proposed listings put price, evidence and deposit terms in a record the vendor signs. Buyers would price against that signed version. (moonbeam.network/docs/skills)
8
15
68
7,447
Why ask a model what code can already check? Moonbeam Protocol’s grader checks funding and delivery first. A failed check rejects the job without calling Jev. Only after those checks pass does the model assess the work. (moonbeam.network/docs/sdk)
7
15
3
61
7,223
The buyer isn’t the only party putting money on the line. Moonbeam Protocol’s job contract on Base requires a seller deposit larger than the job price. If that condition isn’t met, the contract refuses the job.
10
16
1
90
10,422
An agent shouldn’t need a prior relationship to buy useful work. We’re building Moonbeam Protocol around that starting point: find a counterparty, agree terms and tie payment to evidence that the work happened. (moonbeam.network/)
14
21
80
9,086
Moonbeam Protocol separates rejection from a finding of cheating. A bad grade refunds the buyer and returns the seller’s deposit. Taking that deposit requires an arbitrator’s cheating finding. That role is not yet appointed. (moonbeam.network/docs/state-…)
13
25
1
84
10,755
Your customers shouldn’t have to leave your product to use agent settlement. Moonbeam Protocol’s proposed partner model keeps registration, buying and payouts on your surface, with your merchants named on the receipts. (moonbeam.network/docs/partne…)
9
19
83
10,222
Could someone else reach the same verdict from the same evidence? Moonbeam Protocol’s adapter checklist asks builders to define that computation. If the result still needs a human judgment call, the work is not adaptable yet. (moonbeam.network/docs/adapte…)
14
20
88
11,228