Open Grant Proposal: Echo — Universal AI Context Portability on Filecoin FEVM

[Mayorken]

Open Grant Proposal: Echo — Universal AI Context Portability on Filecoin FEVM

Proposal Category: FVM / Integrations / Developer Tooling

Proposer: Kenneth Umoekpe (kennethumoekpe@gmail.com)

GitHub Repository: https://github.com/Mayorken/Echo

Do you agree to open source all work you do that receives Filecoin grant funding? Yes

Project is: Active (deployed to Calibration testnet, full test suite passing, live demo running)

Recognition: 1st Place, FilecoinTLDR Builder Challenges Cycle 1 (announcement)

Note: This is a revised, smaller-scope version of a proposal previously posted in this repository. The original asked for $50K across four milestones including an external audit, full hosted gateway, published docs/npm package, and a monitoring stack. This version narrows the ask to the single highest-leverage milestone — mainnet deployment plus the minimum Web2-friendly onboarding needed for non-crypto-native developers to actually use Echo without touching a wallet.

---

Project Description

The Problem

Every developer using AI tools pays an invisible tax: when you switch tools, you start over.

A developer builds months of context with a coding assistant — their stack preferences, architecture decisions, naming conventions, how they want code explained. All of it lives inside one company's servers. Switch to a better model, get your account suspended, or have the platform change its infrastructure — and everything is gone.

Nobody has fixed this because no centralized company has an incentive to. Lock-in is a feature for them, not a bug. Attempts to solve this with a centralized database just move the trust problem — whoever runs the database becomes the new gatekeeper.

The Solution

Echo is a protocol layer that sits underneath AI tools rather than being one itself. It treats AI context the same way OAuth treats identity — as a portable, user-owned primitive that any tool can access with permission and no tool can hold hostage.

The architecture is simple:

1. A user's context is encrypted on their device (AES-256-GCM) and stored on Filecoin via the Synapse SDK
2. A smart contract on FEVM holds the CID pointer and access list — the user controls which AI tools can read their context
3. Any AI tool with the right ABI and a granted address can call getMemory() and load full context instantly
4. An auto-renewal keeper monitors CIDs and re-pins expiring Filecoin deals from a pre-funded on-chain endowment

Switch from Claude to Gemini to Codex mid-project — the new tool picks up exactly where the last one left off. No re-explaining. No starting over. And because the storage lives on Filecoin rather than inside any AI company, no platform can take it away.

What Is Already Built

Echo is not a sketch. Every component below is deployed and working:

Component	Status
EchoMemoryRegistryV3.sol — UUPS-upgradeable FEVM smart contract	Deployed: 0x962C42f208d89D5bF1698E3397BC78176D70cE0c (Calibration)
Team Vaults — on-chain RBAC shared context for engineering teams	Complete
Keeper spend path — keeperDeductRenewal(), CEI + nonReentrant	Complete
echo-sdk.js — JavaScript SDK with real AES-256-GCM encryption	Complete
Synapse SDK storage adapter — real Filecoin deals on every saveMemory()	Complete
Auto-renewal keeper bot — scans events, checks deal status, re-pins	Complete
REST API — 13 endpoints, rate-limited, helmet-hardened	Complete
MCP server — 13 tools for Claude Desktop (stdio JSON-RPC transport)	Complete
OpenAPI 3.0 spec — ChatGPT Actions compatible	Complete
Funding bridge — Stripe webhook → fundRenewal() on-chain	Complete
Internal security review — 0 critical/high, 5 medium/low found and remediated	Complete
Live demo — npm run demo, 6 scenarios, real chain, no external deps	Complete

The contract has been live on Calibration testnet since June 21, 2026. Running npm run smoke executes live on-chain checks against it in seconds.

Why Filecoin Specifically

The portability promise only holds if the underlying storage is genuinely permanent and user-controlled.

Filecoin's perpetual-storage mechanism means a user funds a one-time endowment and Echo's keeper keeps renewing the storage deal indefinitely — no subscription, no company decision to reverse. Proof of Data Possession means the integrity of stored context is mathematically verifiable, not just claimed. And programmable access control via FEVM means the user's permissions aren't a policy someone could quietly change — they're code running on a public network.

Echo creates real, ongoing Filecoin storage demand. Every saveMemory() call makes a new Filecoin deal via the Synapse SDK. Every keeper run renews expiring deals. As AI tool adoption grows, so does the FIL and USDFC locked in storage — permanently.

---

Value to the Filecoin Ecosystem

Demand-side storage: Each Echo user generates multiple Filecoin storage deals over their lifetime via Synapse. A user who switches between three AI tools per week creates a new deal on each context save and renewal — genuine paid on-chain deal demand.

Three AI platforms connected to Filecoin storage today: The MCP server connects Claude Desktop users to Filecoin. The OpenAPI spec connects ChatGPT users. The REST API connects every other AI tool with HTTP capabilities.

The core bottleneck to adoption is Web2 friction, not the protocol. A developer who has never touched a wallet won't self-host a keeper or manage a private key just to try Echo. This grant funds exactly the piece that removes that friction: social login instead of a seed phrase, and a hosted API instead of self-hosting.

Novel FVM use case: On-chain access control for off-chain AI context pointers is a new primitive. The keeper spend path — where an authorized address pulls reimbursement from a user's on-chain balance after servicing their storage deal — demonstrates a self-sustaining economic model built entirely on FEVM.

---

Deliverable: Mainnet Launch + Web2-Accessible Onboarding

Everything in this single milestone is scoped around one goal: someone who has never used a wallet can sign in with Google and have working, portable AI context on real Filecoin storage in under a minute.

• Deploy EchoMemoryRegistryV3 to Filecoin mainnet behind ERC1967 proxy
• Transfer contract ownership to a Gnosis Safe multisig (removes single-key admin risk)
• Verify contract source code on Filfox block explorer
• Fund and configure a Synapse SDK mainnet wallet; deploy the keeper bot pointed at mainnet storage (basic scheduled run — no monitoring dashboard in this phase)
• Web3Auth social login — Google / GitHub sign-in, no wallet or seed phrase required, non-custodial key derivation
• API key auth on the REST API — hosted instance behind standard API key auth so developers get a key on signup instead of self-hosting
• Minimal user dashboard — a single page showing stored context, which tools have access, and renewal balance (no payment UI in this phase — the Stripe webhook bridge already exists in the codebase and can be wired to a checkout flow in a follow-up ask once there's real usage to justify it)

Explicitly out of scope for this ask (deferred to a future proposal once the project has mainnet traction to point to): a formal paid external audit (Cyfrin/Sherlock-tier, $10-30K on its own), Grafana/PagerDuty monitoring infrastructure, a published npm package with a full documentation site, and integration sample repos/video walkthrough. The internal security review already covers the contract to a reasonable bar for an initial mainnet launch at this scale; a paid external audit is the right next ask once there's real usage and TVL to justify that spend.

Duration: 4-5 weeks
Budget: $15,000

Task	Detail	Est.
Mainnet contract deployment	npm run deploy against mainnet RPC; verify addresses
Gnosis Safe setup	Deploy Safe on FEVM mainnet; transferOwnership() to Safe address	~$3,000
Contract verification	Submit source to Filfox; confirm bytecode match
Synapse mainnet wallet + keeper launch	Fund wallet with FIL + USDFC; configure keeper with SYNAPSE_PRIVATE_KEY; confirm first mainnet sweep	~$2,000
Web3Auth social login	Integrate @web3auth/modal; Google + GitHub providers; returns an ethers.Signer to the SDK	~$4,000
API key auth middleware	Add API key generation + validation to the REST API, scoped per user wallet	~$2,000
Minimal hosted dashboard	Single-page view: stored context, access list, renewal balance; deployed with hosting	~$4,000

Verifiable completion criteria:

• Contract visible on Filfox mainnet explorer with verified source
• owner() on mainnet proxy returns a Gnosis Safe address
• Keeper logs show at least one successful mainnet sweep
• A user can sign in with Google (no wallet), save and load context, and see it reflected in the dashboard — entirely without touching a private key

---

Team

Kenneth Umoekpe — Project Lead and Primary Developer

• Built the complete Echo infrastructure described in this proposal: EchoMemoryRegistryV3.sol, echo-sdk.js, keeper bot, REST API, MCP server, OpenAPI spec, funding bridge, and internal security review
• Hands-on across Solidity, JavaScript/Node.js, ethers.js v6, Filecoin FEVM, the Synapse SDK, and the MCP protocol
• 1st Place, FilecoinTLDR Builder Challenges Cycle 1 (announcement)
• Repository: https://github.com/Mayorken/Echo

The codebase is open source and fully reviewable. Run npm run demo to see every component working end-to-end on a local chain in under 2 minutes, or npm run smoke to verify the live Calibration deployment in seconds.

---

Risks and Mitigations

Risk	Likelihood	Mitigation
Deferring the external audit before mainnet is risky	Low-Medium	Internal review already found and fixed the only medium/low findings; contract uses CEI pattern, nonReentrant, and zero-address guards throughout. A formal external audit is planned as the next ask once mainnet usage justifies the spend.
Web3Auth social login has UX friction on FEVM	Medium	Web3Auth supports custom EVM chains; Calibration and mainnet are EVM-compatible. Fallback: private key import for developers who prefer it.
Synapse mainnet storage costs exceed expectations	Low	USDFC-denominated pricing is predictable and small at this scale; the keeper fee mechanism already accounts for re-pinning costs.
Low initial developer adoption	Medium	MCP integration means every Claude Desktop user is one config block away from Echo. Social login removes the single biggest onboarding blocker for non-crypto-native developers.
Single-developer team creates bus-factor risk	Medium	All code is open source, fully documented, and structured for handoff. The single milestone here is concrete and independently verifiable.

---

Additional Information

Live demo: https://echofmemory.netlify.app

Contract on Calibration testnet:
0x962C42f208d89D5bF1698E3397BC78176D70cE0c
Verify live: npm run smoke

Quick local verification:

git clone https://github.com/Mayorken/Echo
cd Echo
npm install
npm run demo      # 6 scenarios, real chain, ~30 seconds

Why now: The AI tool market is consolidating around 3-4 dominant platforms (Claude, ChatGPT, Gemini, Copilot). The window to establish context portability as a protocol-level primitive — before each platform builds its own walled garden solution — is the next 12-18 months. This scaled-down ask gets Echo onto mainnet and removes the wallet-friction barrier for Web2 developers now, with the larger deliverables (audit, monitoring, docs) as natural follow-on asks once there's real usage to point to.

Grant contact: grants@fil.org / #grants-help on Filecoin Slack

[Mayorken]

Thanks for the interest, but I don't think this is applicable to Echo's current architecture.

The "main" field in package.json already resolves correctly to echo-sdk.js, which is the public entry point re-exporting EchoClient, generateEncryptionKey, and createLighthouseStorage. Since the project is pure CommonJS JavaScript (no TypeScript compilation step), there's no emitted output directory or .d.ts file that could be "missing" the source is the distributable.

An "exports" map would only be relevant if we were publishing to npm with subpath exports (e.g. echo-context/storage, echo-context/crypto), which isn't the current distribution model consumers clone the repo or point integrations at the SDK directly.

If you're seeing a specific resolution failure when requiring the package, please share the exact error and the Node.js version you're testing against. Otherwise I don't see an actionable issue here — npm pack --dry-run correctly includes echo-sdk.js as the entry point and all submodules under lib/, integrations/, and keeper/ are accessible via relative paths as documented.