Protect execution INTENT (pre-mempool) in addition to v0.1's metadata + local-data privacy, via a pluggable private-routing layer behind the existing execution engine.
Research + prototype scaffolding only. No signing, relaying, broadcasting, or execution is implemented.
Final on-chain settlement stays publicly visible on Base. Full on-chain privacy is NOT claimed.
Architecture (pluggable, behind the execution engine)
1. Copilot Intent — User expresses a goal (deposit/withdraw/swap). Already exists.
Sanitize into an abstract RouteIntent (size buckets, opaque target refs) — no raw secrets.
2. Privacy Router (abstraction) — Select a routing strategy + build a read-only plan; single seam the rest of the app calls.
Provider interface has NO execute() — execution stays in the audited execution engine, gated by flags.
3. Provider Adapters — private-rpc / mev-protected / intent-based / relayer / stealth-wallet / encrypted-orderflow / privacy-network.
Each adapter is independently audited + feature-flagged; default OFF.
4. Existing Execution Engine — User confirmation + wallet signature (unchanged). Only the SUBMISSION transport changes per provider.
No new signer logic; consent + signing remain exactly as today.
5. Settlement (Base) — On-chain inclusion + settlement.
Public regardless of routing — this is the honesty boundary.
Approaches ranked by research privacy score (read-only preview)
encrypted-orderflowprivacy 88/100 · mempool strong · MEV strong
Cost +$0.1–1 · sender-unlinkability partial · amount strong
intent-basedprivacy 75/100 · mempool strong · MEV strong
Cost +$0–0.2 · sender-unlinkability partial · amount partial
privacy-networkprivacy 75/100 · mempool strong · MEV partial
Cost +$0.1–1.5 · sender-unlinkability strong · amount partial
private-rpcprivacy 50/100 · mempool strong · MEV strong
Cost +$0–0.05 · sender-unlinkability none · amount none
mev-protectedprivacy 50/100 · mempool strong · MEV strong
Cost +$0–0.1 · sender-unlinkability none · amount none
relayerprivacy 38/100 · mempool partial · MEV partial
Cost +$0.02–0.3 · sender-unlinkability partial · amount none
stealth-walletprivacy 25/100 · mempool none · MEV none
Cost +$0.05–0.5 · sender-unlinkability strong · amount none
Phased rollout
Phase 1: Private RPC + MEV protection low effort
Keep txs out of the public mempool and stop front-running — near-zero cost, no UX change.
UX: Optional 'Protected submission' toggle; otherwise invisible.
Phase 2: Intent-based execution medium effort
Route swaps via solver/auction networks to obscure path + protect against MEV.
UX: Swaps may settle via batch auction; better price + privacy.
Phase 3: Relayer + stealth agent wallets high effort
Decorrelate the funding wallet from on-chain actions and unlink agent history.
UX: Agent wallet becomes a smart account; per-action stealth addresses (opt-in).
Phase 4: Encrypted order flow + privacy network high effort
Hide intent AND amount pre-execution via threshold/TEE encryption or a dedicated privacy network.
UX: Strongest pre-execution privacy where infrastructure exists on Base.
Tradeoffs
• Privacy vs latency: Encrypted/threshold + auctions add inclusion latency vs a direct private RPC.
• Privacy vs cost: Stealth wallets + encrypted flow add gas/fees; private RPC is ~free.
• Privacy vs decentralization: Relayers/builders/solvers/committees introduce trust + censorship points.
• Privacy vs UX: Account migration, stealth bookkeeping, and recovery add user complexity.
• Pre-execution vs settlement: All approaches still settle publicly on Base — they hide intent, not the final record.
Cost (research estimates / action)
• Private RPC: ~$0 (sometimes rebates)
• MEV-protected: ~$0–0.10 (often net-positive)
• Intent-based: $0–0.20 (often offset by price improvement)
• Relayer (4337): $0.02–0.30 (bundler/paymaster premium)
• Stealth wallets: $0.05–0.50 (fund + sweep overhead)
• Encrypted order flow: $0.10–1.00 (early infra)
• Privacy network: $0.10–1.50 (provider-dependent)
Security
⚠ Every transport (RPC/relayer/builder/solver/committee) is a trust + liveness point — always provide a clearly-labeled fallback.
⚠ Never let any provider construct calldata or move funds without the user's explicit signature (consent stays in the execution engine).
⚠ Stealth + encrypted approaches add key-management risk — design recovery before enabling.
⚠ Audit each adapter independently; ship behind a default-OFF feature flag.
⚠ Be explicit in-UI that settlement is public; never imply full anonymity.
User flow — default vs private routing
Default:
1. User: 'Deposit 1,000 USDC into the best vault'
2. Copilot builds the action preview (as today).
3. User confirms → wallet signs → tx broadcasts via the normal RPC → public settlement.
Private routing (opt-in):
1. User enables 'Protected submission' (Private Mode → routing).
2. Copilot builds the SAME action preview + a read-only ROUTE PLAN (provider, privacy score, est. cost, trust notes).
3. User confirms → wallet signs the SAME tx/intent → submitted via the chosen private transport (e.g., private RPC) instead of the public mempool.
4. On transport failure, Bloom shows a privacy-downgrade notice before any public fallback.
5. Settlement is still public on Base — the UI says so plainly.
Recommendation
Phase 1 (Private RPC + MEV protection): highest privacy-per-effort, ~free, no consent/signing changes, easy fallback.
Phase 2 (intent-based) for swaps, then evaluate Phase 3/4 as Base infra matures.
Keep execution in the audited engine; private routing only changes the SUBMISSION transport, never the signing/consent model.
Research + prototype scaffolding only. No execution is implemented; the planner's execute path throws by design. On-chain settlement remains public on Base.