The dual-score podcast for decentralised AI. Every episode breaks down a DeAI project with honest, data-driven assessments, scoring both sovereignty (Freedom Score) and token economics (Returns Score). No sponsors, no shilling, no hype.
Reading the feed…
FLock brings NeurIPS-awarded federated learning to a blockchain, with real protocol revenue from a small number of training tasks. It grades C on freedom and C on returns. We cover privacy-preserving training, the token, and whether a handful of curated tasks becomes thousands.
Cookie DAO is a legitimate AI-agent analytics product with an industry-cited Mindshare metric and first-mover advantage. It grades F on freedom and D on returns. We cover the agent-analytics thesis and why the 'DAO' label is marketing: no open-source code, no verifiable governance, centralised infrastructure.
Allora is a self-improving inference network with genuine technical innovation and tier-1 exchange listings from day one. It grades D on freedom and D on returns. We cover the regret-based aggregation, the unproven revenue model, and a large insider cliff in late 2026.
Morpheus, NEAR and Venice all let you stake a token and draw AI inference against it, and all three pay the compute bill out of issuance rather than customer fees. We cover what each entitlement is pegged to, why the consolidated balance sheet flips the ranking, and what would make the model fund itself.
OriginTrail is the quiet achiever of DeAI, a decentralised knowledge-graph protocol turning data into verifiable assets for AI and enterprise. It grades C on freedom and C on returns. We cover the knowledge graph, arguably the best tokenomics in the sector, and whether provenance is a real AI use case.
Nillion has the right thesis, that AI needs privacy-preserving computation, built on multi-party computation and secure enclaves. It grades C on freedom and D on returns. We cover blind compute, the completed migration to Ethereum, and whether it can scale to real workloads after a rocky market-maker episode.
Vana is arguably the most technically complete data-sovereignty protocol in crypto, where users contribute personal data to DataDAOs and earn from AI training. It grades C on freedom and C on returns. We cover the data-DAO model, the permissioned validators, and whether users can really monetise their own training data.
Theta is enterprise-backed decentralised cloud with Google, Samsung, and Sony among its validators. It grades C on freedom and D on returns. We cover the edge-compute pivot, and THETA's fundamental value-accrual problem: the fees flow to its TFUEL gas token, not THETA.
NEAR has the strongest AI credentials of any Layer-1 founder, a co-author of the transformer paper, on a proven sharded chain with real cross-chain traction through Intents. It grades C on freedom and B on returns. We cover the sharding, the AI roadmap, the activated fee switch, and whether the infrastructure thesis holds.
Oasis is a privacy-first L1 with confidential smart contracts, finally shipping product like off-chain compute mainnet and an institutional pilot. It grades C on freedom and D on returns. We cover the Sapphire ParaTime, ROSE tokenomics, and why AI is now a strategic focus.
The 5-level privacy spectrum for AI inference. Anonymisation vs confidentiality. What Venice actually delivers, what Morpheus TEE integration means, and how to choose private inference.
Phala is a confidential-computing cloud built on hardware secure enclaves, with enterprise compliance credentials. It grades C on freedom and D on returns. We cover how the enclaves keep data private even from the provider, the private-inference tie-in, and what still has to scale.
ElizaOS built the open-source AI agent codebase other projects wish they had and a token nobody can value. Virtuals built strong token mechanics on a centralised inference engine. Same agent category, opposite trade-offs.
There is no RENDER staking. What actually exists: GPU node operation, availability rewards, burn-mint emissions, and the honest economics of providing compute to the Render Network.
TAO is the token. Subnets are the business. Which Bittensor subnets earn revenue from paying customers, which ones are farming emissions, and how to tell the difference. Chutes, Gradients, Templar and the Rayon Labs concentration question.
Akash's BME activation changes everything for AKT holders. How burn-mint works, the Render comparison, what the first week of metrics tells us, and whether BME fixes AKT's inflation problem.
Ocean pioneered decentralised data markets, and its Compute-to-Data model remains one of the most elegant privacy solutions in Web3. It grades C on freedom and F on returns. We cover data tokens, compute-to-data, the ASI Alliance merger terms, and a Foundation that says the token has no intended utility.
What the FET/AGIX/OCEAN/CUDOS merger actually means for token holders. Conversion ratios, Ocean's exit and lawsuit, the merged supply dynamics, and whether the alliance is more than branding.
Virtuals is the dominant AI-agent launchpad, now spanning five chains, with on-chain revenue and a buyback-and-burn programme. It grades D on freedom and C on returns. We cover the GAME framework, agent tokens, revenue sharing, and where value actually accrues.
Morpheus pays its inference providers from a protocol emission, not from what users spend, and that single fact decides whether the market is sustainable. A trace of who pays, who earns, and why the staked path returns your MOR in full while only the direct-pay path spends it. Plus what our own on-chain indexing shows: a working marketplace at a small scale, a narrow set of providers, and demand still mostly subsidised by emissions rather than paid by users.
The word agent now covers chatbots, trading bots, launchpad tokens and robot runtimes, and they earn opposite verdicts. This is a map, not a buyer's guide. Frontier agent modes from the big labs set the usefulness bar, and most tokenised agents sit below it. Where a specialist actually earns its keep, why sovereignty only matters the day an operator turns hostile, and the one test that sorts the field: would you still run it if its token went to zero.
How Bittensor's dynamic TAO actually works. Alpha tokens, flow-based emissions, AMM mechanics, and the zero-emission death spiral risk for subnets that don't produce value.
Morpheus builder subnets are a public goods funding mechanism with the same design as Gitcoin Grants. There are three reasons to stake MOR, yield, access and conviction, and each does something different to the token. What the design intended, what each choice does to MOR, and the Gitcoin protections that keep the healthy paths dominant.
Bittensor's first halving and what it means for TAO holders. Emission reduction, the dTAO interaction, and whether the Bitcoin scarcity playbook applies to a network funded by inflation.
Gensyn is verifiable AI compute, proving ML work happened correctly. It grades D on freedom and F on returns. We cover the proof system, why verification matters, and the pre-token risk profile.
