Standard Grant Proposal: SiaHost Agent

Alignment with April 2026 Grants Program Guidelines

This proposal fully complies with the Sia Foundation’s new funding directives effective April 10, 2026. It is explicitly focused on building on indexd (real-time network data ingestion + hostd signals for autonomous host optimization).

The project directly advances the Foundation’s stated push for robust tooling built on indexd, as called out in the new guidelines on Global Grants Program | Sia. All other requirements (open-source, measurable milestones, high-level architecture overview, risk mitigation, and monthly reporting per the official Grants Development Guide) are addressed in detail below.

Introduction

Project Name: SiaHost Agent — Autonomous Host Optimization System
Name of the organization or individual submitting the proposal: Dapp Mentors


Background

Dapp Mentors is a two-time Sia Foundation grant recipient and an AI + blockchain development firm specializing in agentic automation, distributed systems, and developer tooling. We build infrastructure-grade solutions that simplify decentralized technologies. Previous Sia-funded work includes SiaLearn, a developer education program with video tutorials, hands-on projects (such as VidTV using Sia Renterd), and SiaPeopleLearn, a decentralized educational platform built on Sia Renterd for hosting courses and content.

Describe your project

SiaHost Agent is a local-first, non-custodial, agentic automation system built on indexd and hostd. It combines real-time network data from indexd with hostd signals, using an event-driven architecture, LangGraph + Claude Sonnet AI agents, Temporal workflow orchestration, and n8n automation to continuously optimize pricing, collateral allocation, uptime, and contract selection directly on the host machine.

The system ingests host and network signals, reasons over them intelligently, and executes safe, controlled optimizations via the MCP Tool Server. By abstracting the technical burden of host management, SiaHost Agent enables both technical and non-technical users to run high-performing storage nodes with minimal effort.

A full visual and architectural breakdown of the 9-layer system design, data flow, and autonomous optimization lifecycle is available here:

This agentic layer transforms Sia hosting into a self-optimizing system, boosting reliability, competitiveness, and overall network efficiency.

Who benefits from your project

The primary beneficiaries are Sia host operators, particularly those without deep technical expertise. By reducing operational complexity, more users can participate in the network.

The Sia network benefits through improved uptime, better pricing equilibrium, and increased storage supply. Renters benefit indirectly from more reliable hosts and competitive pricing.

Developers and ecosystem contributors benefit from an extensible open-source framework that can be expanded for further automation tools.

By enabling non-technical operators to run competitive hosts, SiaHost Agent is projected to meaningfully increase decentralized storage supply and uptime across the network — directly supporting the Foundation’s mission of user-owned data at scale.

How does the project serve the Foundation’s mission of user-owned data? What problem does it solve?

The project removes technical barriers that prevent broader participation in decentralized storage infrastructure. Running a competitive Sia host currently requires manual tuning of pricing, collateral, and performance strategies.

SiaHost Agent solves this by introducing intelligent automation that ensures hosts remain competitive and reliable without requiring deep expertise. This increases decentralization by enabling more independent operators to participate.

By deeply integrating with indexd for network-wide insights, the project directly advances the Foundation’s push for robust tooling built on indexd, while increasing decentralized storage participation.

Compliance Questions
Are you a resident of any jurisdiction on that list? No
Will your payment bank account be located in any jurisdiction on that list? No


Grant Specifics

Amount requested and budget breakdown
Total Requested: $38,000 USD
Duration: 5 Months
Monthly Allocation: $7,600 USD (Equal Distribution)

Category Monthly Cost ($) 5 Month Total ($) % of Budget Budget Justification and Primary focus
Core Engineering & AI Development 4,600 23,000 60.5% Months 1–3: FastAPI backend, hostd/indexd integrations, MCP Tool Server, OptimizerAgent, LangGraph + Claude Sonnet logic, pricing/collateral engines, Temporal workflows, n8n automation
Frontend Development (Dashboard UI/UX) 900 4,500 11.8% Month 4: Next.js 16 dashboard, real-time metrics, onboarding wizard, shadcn/ui components, additional UI/UX fine-tuning.
Infrastructure & DevOps 700 3,500 9.2% Months 1 & 4: Docker Compose, PostgreSQL + pgvector, Qdrant, Temporal/n8n setup, Prometheus/Grafana/Loki observability stack
Testing, QA & Security 700 3,500 9.2% Month 5 + ongoing: E2E testing, RBAC, audit logs, signed actions, performance tuning, failure handling
Documentation & Open Source Release 400 2,000 5.3% Month 5: Technical documentation, GitHub repo finalization, community materials
Project Tools & Miscellaneous 300 1,500 3.9% Project tools licensing, Emergency funds across all months (absorbed where possible into core work)
Total 7,600 38,000 100%

High-level architecture and security practices

The system follows a local-first, event-driven modular architecture deployed as a self-hosted Docker Compose stack. All decision-making agents and workflows run locally without exposing private keys. Secure integration with hostd and indexd occurs via the MCP Tool Server. The 9-layer design includes data ingestion, AI intelligence (LangGraph + Claude Sonnet with Qdrant vector memory), durable workflow orchestration (Temporal + n8n), FastAPI backend, Next.js dashboard, PostgreSQL/pgvector persistence, and full observability.

Security best practices: non-custodial design, sandboxed execution, signed actions, RBAC, audit logs, encrypted communication, strict separation of decision logic and execution layers, and fail-safe defaults with manual override capability.

The complete 9-layer architecture (data ingestion → AI intelligence (LangGraph + Claude Sonnet + Qdrant) → Temporal/n8n orchestration → FastAPI backend → Next.js dashboard → PostgreSQL/pgvector persistence + full observability) follows the exact structure recommended in the official Sia Foundation Grants Development Guide.

A visual diagram and interactive 3D user journey map are embedded in the linked GitHub repository and Figma document (see Development Information section). All decision-making remains local-first and non-custodial with no private keys exposed.

Timeline and Milestones

Milestones follow the PR-driven workflow per the Sia Foundation Grants Development Guide. Each task below maps to one focused pull request with a clear title, description, and testing instructions. Monthly reports will link all completed PRs via the official report template.

Month 1 — Foundation + Data Layer

Task PR Scope
Repo setup, CI/CD, Docker Compose base config Infrastructure scaffold
PostgreSQL + pgvector schema + migrations DB layer
Temporal + n8n bootstrap with health checks Orchestration init
FastAPI skeleton: routes, Pydantic models, error handling API skeleton
hostd API client: connect, authenticate, poll metrics hostd integration
Metrics ingestion pipeline → PostgreSQL Data ingestion

Month 2 — Intelligence Layer

Task PR Scope
indexd API client: stream network pricing + contract data indexd integration
MCP Tool Server: core tools (get_host_stats, get_network_state, propose_price_change) MCP server base
Rule-based pricing + collateral engines Optimization rules
Qdrant vector store for agent memory Vector memory
OptimizerAgent v1: LangGraph + MCP tools (rule-based) Agent v1
OptimizerAgent v2: Claude Sonnet reasoning layer added Agent v2 with AI

Month 3 — Workflow Orchestration

Task PR Scope
Temporal: PricingOptimizationWorkflow Pricing workflow
Temporal: CollateralRebalancingWorkflow Collateral workflow
Temporal: UptimeMonitorWorkflow Uptime workflow
Dead-letter queue + retry policies Failure handling
n8n event triggers + alert routing Event automation
Fail-safe manual override flag Safety override
Full lifecycle simulation test (mock hostd/indexd) Integration test

Month 4 — Operator Control Interface + Observability

Task PR Scope
Next.js 16 + shadcn/ui scaffold + routing Frontend scaffold
Onboarding wizard: connect hostd, set risk tolerances Onboarding flow
Agent control panel: approve / reject / override decisions Control panel
Actionable alerts panel with one-click response Alerts UI
Prometheus metrics export from backend + workers Metrics export
Grafana dashboards: agent health, decision latency, errors Grafana setup
Loki log aggregation across all services Log aggregation

Month 5 — Hardening, Testing & Launch

Task PR Scope
RBAC: role definitions + permission middleware Access control
Audit log: immutable record of all agent decisions Audit logs
Signed actions: cryptographic signing before execution Signed actions
E2E test suite: full optimization lifecycle coverage E2E tests
Performance tuning: decision loop latency + DB queries Perf tuning
README + technical docs: build/run/config instructions Documentation
Public launch: Forum post, Discord, GitHub v1.0.0 release Open-source release

Evidence of Demand

The shift toward autonomous, hands-free infrastructure is no longer emerging — it is here. Gartner projects 40% of enterprise applications will embed task-specific AI agents by end of 2026, up from less than 5% in 2025. Gartner

Deloitte reports 93% of IT leaders plan to introduce autonomous agents within two years. OneReach Within decentralized infrastructure specifically, the DePIN model is rapidly gaining ground — but its success depends entirely on independently operated nodes staying competitive, reliable, and well-configured. Medium

Most Sia host operators today rely mostly on manual hostd tuning, static scripts, or community guides — approaches that limit non-technical users and struggles to fully adapt to live network conditions. No open-source, AI-driven optimization system currently exists for Sia hosts. SiaHost Agent fills that gap directly, and in a machine economy where competing infrastructure is increasingly automated, a manually operated host is at a structural disadvantage.

Target user

Primary: Developers and technically capable host operators building on the Sia ecosystem — the open-source, extensible framework is designed to be forked, extended, and integrated into broader infrastructure tooling. The MCP Tool Server, LangGraph agent architecture, and modular pipeline design are all developer-friendly extension points.

Secondary: Independent Sia host operators of all technical levels — the system’s automation layer abstracts operational complexity, enabling even non-technical node runners to maintain competitive, well-configured hosts without deep expertise.

Potential risks

  • AI hallucination / incorrect decisions → Mitigation: Use of high-grade agentic models such as Claude Sonnet 4.6, combined with systematic prompt optimization through an Opik dashboard. This approach reduces hallucination risks to near-zero while maintaining robust decision quality. Additional safeguards include rule-based constraints, and extensive simulation testing before live execution.
  • Sia hostd / indexd API changes → Mitigation: Adapter abstraction layer
  • Workflow failures → Mitigation: Temporal durability + dead-letter queues
  • Data inconsistency → Mitigation: Event sourcing and audit logs

What are your plans for this project following the grant?

Following the grant, SiaHost Agent will continue as a fully open-source, community-maintained project with no plans for monetization. All future development will be driven through the public GitHub repository in collaboration with the Sia ecosystem.

Post-Grant Technical Roadmap:

  • v1.1 (Month 6–7): Community feedback integration, hostd/indexd compatibility updates (IF ANY) for any new releases, expanded optimizer rule sets based on real host operator data.
  • v1.2 (Month 8–10): Multi-host support for operators managing multiple nodes, plus RAG-powered chat with the SiaHost Agent for direct, context-aware information.
  • v2.0 (Month 11–13): Extended SDK integrations as the Sia Dev Portal matures, potential cross-platform packaging (standalone binary distribution), and expanded AI agent capabilities based on live network learnings.

Community Channels: We will maintain active project presence on:

  • GitHub Discussions — primary channel for technical Q&A, feature requests, and contributor coordination
  • Sia Discord — ongoing presence in the grant channel for community support and feedback
  • Sia Forum — monthly progress reports and post-launch update threads
  • X/Twitter & LinkedIn — project announcements and ecosystem visibility

Sustainability Plan: SiaHost Agent is designed to remain free and open-source. Future sustainability will be supported through: (1) future Sia Foundation grant applications for major version upgrades, (2) Dapp Mentors’ continued organizational commitment to the Sia ecosystem as a long-term contributor. The project will never require a paid tier or token.


Development Information

Open-source status
Yes. All core components will be open-source. Any third-party dependencies will comply with open-source licensing standards.

Leave a link where code will be accessible for review

All project assets, documentation, and implementation details are maintained within the repository. The architecture blueprint and user journey flow can be accessed at:

Monthly progress reports
Yes, provided in the Sia forum and also on the GitHub repo change-logs.

Monthly reports will use the official template at Monthly Grant Report Template | Sia, with direct links to pull requests for every completed task.


Contact Info

Email:
[email protected]
Other contact methods:
LinkedIn: https://www.linkedin.com/in/darlington-gospel

@DappMentors - thank you for your proposal.

Your proposal comes at an interesting time, however, as we will be releasing new Grants Program funding guidelines this week. This also means the next Grants Committee meeting will be held on April 28th (with April 22 as the proposal submission deadline) to allow for adequate time for these new guidelines to be reviewed & incorporated into proposals.

Please review these guidelines when they’re released in the coming days, and then tag me when/if you’ve updated your proposal accordingly to be reviewed.

Hi @DappMentors - the new funding guidelines have been posted:

After reviewing, please tag me to either look over your edited proposal or to let me know if this proposal should be moved to ‘Inactive.’

1 Like

Hello @mecsbecs
I have made some edits on the main proposal as per your directive and it now fully aligns with the Foundation’s new guidelines.

You’re mentioning the AI agent helping with:

  • Setting up the host → Setting up the host is a one-time task, so setup documentation should be sufficient to guide users through it.
  • Pricing management → This is done by pinning prices to fiat. Unless you’re implying to actively change prices based on host specific utilization patterns. Large pricing changes could negatively affect how renters evaluate a host, and are not something that should be adjusted frequently or without careful consideration.
  • Collateral allocation → The collateral multiplier is the only setting a host operator could use to influence collateral usage (aside from possibly causing contracts to fail by starving available wallet balance). As such, there are no meaningful adjustments an AI agent could make.
  • Monitoring → It’s already possible to receive automated alerts if a host is misbehaving or unresponsive via Siagraph. This feature is also seeing a big update this week. What additional capabilities would an agent provide that are not already possible?
  • Handling updates → Most hosts are installed in the following way:
    • Linux via package repositories → Updates via single command
    • Linux or Windows via Desktop app → Updates automatically
    • Docker → Updates easily when using Docker Compose
    • Custom installed by power users → They likely know what they’re doing and don’t need the help of an agent.
      Updating is already trivial or automated in most cases, and any remaining issues should be addressed through proper migrations.
  • Configuration standardization & Onboarding → This should be handled through setup documentation, defaults, and validation rather than a continuously running agent.

Most of the areas you mention are either already resolved or are one-time configuration problems that proper setup docs should be able to cover.
What specific runtime decisions would the agent make that cannot already be handled statically or through existing tools?

Summarizing the above, one should have gotten some real hosting experience before submitting a proposal on hostd improvements.