Mirrored from filpgf.io — ProPGF Batch 3 (Karma program 1479, application 6a288bbfeaebee9ea5a5c4d7, status: pending). Contact details redacted; canonical application lives on filpgf.io. 1.1 Project Name Mesh Devices: P2P Remote AI Inference (py-libp2p + iroh) 1.2 Project Github https://github.com/acul71/meshdevices 1.3 Project Website https://github.com/acul71/meshdevices 1.4 Team Lead/Point of Contact Luca Pisani — py-libp2p co-maintainer | email: [redacted]| Telegram: @acul71 | GitHub: https://github.com/acul71 1.5 Category [ "Other" ] 1.6 Open Source Status Fully Open Source 2.1 Project Summary RFP: Filecoin ProPGF Batch 3 — Call for Builders (June 2026). Focus Area 3: "AI infrastructure products on Filecoin" — verifiable-compute and audit services, agent audit trails (https://www.filecoin.io/blog/ProPGF-Batch-3-Call-for-Builders). Mesh Devices: dial-by-PeerId remote LLM inference (py-libp2p + iroh) for AI/agent builders. Filecoin stores tamper-evident IPLD inference audit manifests (content-addressed, verifiable retrieval). libp2p handles live streams; Filecoin handles durable provenance. Grant window: 2026-09-01 to 2027-02-28 (6 months). September = grant setup + CI bootstrap; M1–M4 checkpoints through February. Solo part-time. $18,000 total. Releases v0.2.0 and v0.3.0 only — no v1.0. Objective 3 target: enable verifiable AI-inference audit trails stored and retrievable via Filecoin (calibnet paid-deal proof in M4). Lead: Luca Pisani (acul71). 2.2 Who does this work support? [ "Application Builders", "Application Users", "Network Infrastructure" ] 2.3 Total Funding Requested (USD) 18000 2.4 Milestones & Budget [ { "title": "Grant setup + CI bootstrap + baseline demo", "description": "September 2026 (grant month 1): onboarding, milestone plan in repo, CI bootstrap (pytest-trio, ruff), test harness for two-node lm-chat, baseline MVP demo recorded, meshdevices doctor initial version, docs/ grant schedule published. No PyPI release yet — prepares M2 v0.2.", "dueDate": "2026-09-30", "fundingRequested": "4500", "completionCriteria": "- CI workflow runs on main (pytest-trio + ruff)\n- Baseline two-node lm-chat demo recorded and linked in docs/\n- docs/GRANT_SCHEDULE.md (or equivalent) with 6-month plan 2026-09-01 → 2027-02-28\n- meshdevices doctor runs on documented environment\n- Test harness script or docker-compose stub committed for reviewer reproduction" }, { "title": "v0.2 release + cross-NAT hardening", "description": "October–November: Hardened serve/lm-chat UX; identity persistence; NodeTicket cross-NAT documentation; production allowlist enforced. Release v0.2.0 on PyPI with CHANGELOG. Reproducible cross-NAT runbook (docker-compose or script).", "dueDate": "2026-11-30", "fundingRequested": "4500", "completionCriteria": "- Git tag v0.2.0 + CHANGELOG.md; v0.2.0 on PyPI\n- Cross-NAT runbook in docs/ + recorded demo OR reviewer-runnable docker-compose\n- meshdevices doctor passes on documented test environment\n- CI green on main at sign-off" }, { "title": "Backends + IPLD pipeline + v0.3 + tester adoption", "description": "December–January: Pluggable backend (LM Studio + one OpenAI-compatible). Benchmark note. IPLD inference-session manifest spec + example CAR + export script. Concurrent lm-chat clients, systemd example. v0.3.0 on PyPI. GitHub runbook attestation template; recruit ≥2 external testers.", "dueDate": "2027-01-31", "fundingRequested": "4500", "completionCriteria": "- Second backend in CI or runbook; benchmark note in docs/\n- docs/filecoin/ipld-audit-manifest.md + example CAR; export script produces manifest CID\n- Git tag v0.3.0 + CHANGELOG; v0.3.0 on PyPI\n- Recorded demo: two concurrent clients + systemd on Linux\n- ≥2 closed GitHub issues using runbook attestation template\n- ≥1 non-team GitHub issue opened using export script (integration signal)" }, { "title": "Filecoin paid-storage spike + integration guide", "description": "February: One paid storage deal on Filecoin calibration network for M3 IPLD manifest. Demo wallet pays FIL (cost documented). Guide includes tx URL, manifest CID, retrieval command + expected output, payer address. Reviewer-reproducible step-by-step. Calibnet chosen for low-cost reproducible proof; mainnet + deal automation are post-grant non-goals. No v1.0 release.", "dueDate": "2027-02-28", "fundingRequested": "4500", "completionCriteria": "- export script runs in docs walkthrough or CI\n- Calibnet deal/transaction URL(s) in docs/filecoin/int …[truncated] 3.1 Impact pathway Target: 2026 Network Objective 3 — Scale Paid Onchain Flagship Client Adoption (indirect). Downstream actor: AI and agent builders who need tamper-evident inference audit logs and provenance (self-hosted LLMs, multi-agent coordination). Why Filecoin: Content-addressed IPLD manifests + paid storage deals + verifiable retrieval — not achievable with ephemeral P2P streams or local logs alone. RFP Area 3: agent audit trails / verifiable AI infrastructure. Output → Outcome → Impact: - M1 (Sep): CI + baseline — de-risks delivery - M2 (Nov): v0.2 + cross-NAT — builders can deploy P2P inference - M3 (Jan): IPLD pipeline + v0.3 + external testers — reusable audit artifact + adoption signal - M4 (Feb): Paid calibnet deal — onchain storage proof (tx, CID, retrieval, FIL cost) Bridge to Objective 3 KPI: External attestations + non-team export-script usage = leading indicators of flagship-class builder adoption. Paid calibnet deal = verifiable Filecoin retention path agents can follow. Not aggregate flagship client count in this $18K pilot. Calibnet first (low cost, reproducible); mainnet graduation requires follow-on funding (out of scope). Release plan: v0.2.0 (M2), v0.3.0 (M3). No v1.0. 3.2 Verification metrics Objective 3 leading indicators (indirect): 1) Third-party runbook attestation Data source: GitHub issues (attestation template) How measured: ≥2 closed issues from non-team testers (environment, steps, PeerIds) Target: By 2027-01-31 (M3) 2) Export-script adoption signal Data source: GitHub issues/discussions How measured: ≥1 issue opened by non-team user running export script (self-reported) Target: By 2027-01-31 (M3) 3) IPLD manifest pipeline Data source: docs/filecoin/ipld-audit-manifest.md + repo artifacts How measured: Export script produces manifest CID from sample session Target: By 2027-01-31 (M3) 4) Paid calibnet storage deal (M4) Data source: Calibration network explorer + integration guide How measured: Tx URL; manifest CID; retrieval command + output; FIL cost; payer wallet Target: All in docs/filecoin/integration-guide.md by 2027-02-28 Note: Paid onchain deal (demo wallet), not free pin — reviewer-verifiable on-chain 5) Product releases Data source: PyPI + git tags How measured: v0.2.0 (M2), v0.3.0 (M3) — no v1.0 Target: 2 releases 6) CI health at each milestone sign-off 3.3 References - Manusheel Gupta — Filecoin / FHE ecosystem mentor; advised us on ProPGF Batch 3 infrastructure alignment (py-libp2p × Filecoin). GitHub: https://github.com/seetadev - Paul Robinson (pacrob) — py-libp2p maintainer and reviewer; can vouch for our maintainer contributions, interop work, and Filecoin networking context. GitHub: https://github.com/pacrob - Johanna Moran — libp2p/Filecoin ecosystem adviser (pro bono); can speak to our sustained maintainer role across py-libp2p and py-multiformats. Discord: j0hmo 4.1 Monthly Operating Burn [ "< $10K (basic solo operation or part-time team)" ] 4.2 What % of total team monthly burn depends on this grant? 75% 4.3 If this grant is not awarded, what happens? Mesh Devices continues as unpaid part-time work. MVP stays usable locally; v0.2/v0.3 releases, IPLD pipeline, and calibnet proof slip 6–12 months. 75% Note: This refers to Mesh Devices project-focused effort over the grant window, not total personal livelihood. I co-maintain py-libp2p separately (unrelated to this line item). This grant covers the majority of dedicated part-time work on Mesh Devices productization Sep 2026–Feb 2027. Estimated project burn ~$3,000/month (solo part-time labor allocation); $18,000 ≈ 6 months of that focused effort plus ~$500 direct costs in M4. 4.4 Core Team Luca Pisani (acul71) — sole builder, part-time over 2026-09-01 to 2027-02-28. Effort model: sequential milestones M1→M4; September explicitly allocated to grant setup (M1). No paid staff. Budget: $18,000 = M1–M3 labor ($13,500) + M4 labor ($4,000) + M4 direct costs ($500 FIL/tooling). See §2.3 breakdown. Experience: py-libp2p co-maintainer (~627 commits); built Mesh Devices MVP (iroh transport, LM proxy, CLI). Adviser: Manusheel Gupta (Filecoin / FHE). 4.5 Has your team received a ProPGF grant or funding from PLFIF before? [ "No" ] 5.1 Key risks & dependencies - iroh Python API churn — version pinning + IrohTransport abstraction - NAT edge cases — iroh relays, NodeTicket docs, meshdevices doctor - M4 calibnet deal — scoped to one paid deal + guide; not deal automation - Solo builder — public repo, CI, named references mitigate bus factor - Scope — non-goals: v1.0, mainnet production, full deal automation, commercial SLA Anything else you want to share that we didn't ask? Track: ProPGF Batch 3 General Track — RFP Response (Category: Other). RFP: "AI infrastructure products on Filecoin" — Focus Area 3, agent audit trails / verifiable AI infra. Timeline: Start 2026-09-01; M1 checkpoint 2026-09-30 (September work explicit); end 2027-02-28. Objective 3 (indirect): Verifiable AI-inference audit trails on Filecoin — calibnet paid-deal proof + reusable IPLD pipeline + ≥2 third-party attestations. Non-goals: v1.0, mainnet deal automation, full storage client. Repo: github.com/acul71/meshdevices Objective 3 (indirect): Enable verifiable AI-inference audit trails stored and retrievable via Filecoin — leading indicators: ≥2 third-party runbook attestations, ≥1 non-team export-script issue, reusable IPLD manifest pipeline (M3), one paid calibnet storage deal with tx + CID + retrieval proof (M4). Post-grant: mainnet deals and additional builder adopters. Objective 1 N/A Objective 2 N/A Objective 3 Indirect Open Source Context project is fully open source (MIT/Apache-2.0 planned).