Standard Grant Proposal: SiaHost Agent

Grants Proposed
1

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

2

@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.

3

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
4

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.

5

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?

6

Summarizing the above, one should have gotten some real hosting experience before submitting a proposal on hostd improvements.
EDIT: I have tagged you on the Sia Discord. Please engage.

7

@CtrlAltDefeat — Thank you for the precise, point-by-point critique. You correctly identified that one-time setup, static fiat pinning, Siagraph alerts, routine updates, and basic config are already handled well. SiaHost Agent does not duplicate any of them.

It is the dedicated runtime intelligence layer that closes the gaps no static script, alert system, or manual process can sustain 24/7 — built explicitly under the Foundation’s new April 10, 2026 directive: “Building on indexd”.

Here is exactly what the agent does at runtime that nothing else currently handles:

Pricing management You noted that large changes damage renter perception — the agent agrees. It streams live indexd network pricing, contract acceptance rates, and host-score data, then proposes bounded micro-adjustments (±2–5% within your configurable risk bands) only when utilization drops or network supply shifts. This directly addresses the documented 2% hysteresis problem that causes hosts to vanish and reappear on renter allowable lists, and the race-to-the-bottom dynamic seen in the #hosting channel:

  • NickJH (Apr 13, 2026): On the 2% threshold mismatch — “hosts will keep disappearing and reappearing on the allowable hosts lists.”

  • NickJH (Sep 22, 2025): “storage pricing seems to be a race to the bottom… biggest hosts at around 500SC/TB/m… not much chance of making money.”

  • ASPAN (Steve) [JTSK] (Jul 9, 2025): “I have just been looking at the pricing model of the top used storage and there is a theme of low storage price and massive egress pricing… This is why I was becoming a little focused on my own benchmarks.”

SiaHost Agent resolves this by continuously reading indexd signal and proposing signed, constraint-bound micro-adjustments — never large swings, always within the operator’s defined risk band.

Collateral allocation You noted the multiplier is the primary lever. Agreed — but the real-world problem is that actual usage and locked collateral diverge constantly, creating silent renewal failures. The CollateralRebalancingWorkflow monitors actual used storage vs locked collateral, wallet buffer, and indexd contract trends, then proposes safe multiplier or acceptance tweaks before starvation or failure occurs:

  • zipiju (Feb 6, 2026): “the contracted space is 30TB but only 15 is being used… locked collateral is 7KS but storage usage is only 3SC… collateral multiplier of 2, so it should be ~600SC if it actually tracked usage.”

  • Chris (Dec 15, 2025): “We keep running into hosts that fail to refresh and renew contracts due to funding issues… you’ll eventually lose a lot of good data generating revenue.”

  • CtrlAltDefeat (Apr 10, 2026): “If your host grows fast, you’ll need to buy more coins, or risk a big contract not getting renewed.”

  • Javierxam (Mar 12, 2026): Confusion on collateral math — “If I have 10,000 SC risked and a x2 multiplier, may my earnings be something like 5,000SC?”

  • PePeR (Aug 9, 2025): Confusion on renewal payout vs collateral wallet transactions — “I look at numbers and they make no sense to me.”

SiaHost Agent eliminates the need for constant manual wallet monitoring by running this rebalancing continuously, with every proposed change requiring operator approval via the control panel.

Monitoring Siagraph delivers excellent reactive alerts. The agent adds proactive prevention: hostd metrics and indexd network state feed into Qdrant vector memory; Claude Sonnet reasons over emerging patterns; Temporal workflows trigger early interventions before issues escalate to the alert stage:

  • ignited (Jun 20, 2025): “this now happened twice in a week. And even if I don’t notice for a few hours it could be enough to miss a contract renewal or even worse a proof upload window.”

Siagraph tells you something broke. SiaHost Agent acts before it breaks.

Handling updates Package repos, Desktop auto-update, and Docker Compose already handle the mechanics well. The agent layers on safety: pre-update health checks, non-critical workflow pause, post-update metric validation, and automatic rollback if performance degrades after a release:

  • ASPAN (Steve) [JTSK] (Jul 31, 2025): hostd crash with panic — “unexpected contract state transition ‘renewed’ → ‘failed’.”

  • NickJH (Apr 11, 2026): Post instant-sync — “lost nearly all my storage data… none of the big contracts have renewed… my host will be nearly empty this evening” and excessive refreshes draining ~40KS.

  • NickJH (Mar 6, 2026): “28 contract refreshes since 13:27 UTC… hammering my wallet and has drained about 40KS.”

The agent’s post-update validation and circuit-breaker logic would catch the performance regression immediately and halt further action pending operator review.

Configuration & onboarding Documentation and defaults cover the static baseline well. The agent maintains a living optimal config surface informed by real-time indexd network conditions — continuously adapting without requiring manual intervention each time the network shifts.

Summary: Every runtime decision above — bounded micro-pricing, collateral rebalancing, proactive uptime intervention, and post-update validation — requires live indexd data and continuous execution. No static script, documentation set, or alert tool does this. That is precisely the gap SiaHost Agent fills, and precisely what the April 2026 “Building on indexd” directive was written to fund.

All proposed changes are cryptographically signed via the MCP Tool Server and require explicit one-click operator approval. Local-first, non-custodial, 100% open-source.

8

@mike76 — Thank you for the candid feedback and for the direct invitation on Sia Discord: “@Daltonic I’m inviting you to ask the hosts what their pain points are. Without this knowledge, any improvement proposal would be meaningless.”

That was exactly the right challenge. I took it seriously.

Dapp Mentors comes with proven delivery on the renter side and in education (SiaLearn, SiaPeopleLearn) — and that experience gives us a strong foundation for reading the network from the renter’s perspective. For the host side, I went back through 12 months of real #hostd operator messages and grounded every feature of SiaHost Agent in the documented pain points you and others have shared (all quoted and linked in the @CtrlAltDefeat reply above).

Every milestone (Months 1–5) includes live-host validation with experienced operators from this channel — starting with simulation testing in Month 3 and real-host integration from Month 1 onward. We are building with the host community, not around it.

The full 9-layer architecture, PR-driven workflow, risk mitigations, and post-grant roadmap follow the official Grants Development Guide to the letter and are fully aligned with the new “Building on indexd” directive effective April 10, 2026.

Your feedback has already made this proposal stronger. I remain fully open to testing on a real host or jumping on a call at your convenience.

9

Host pricing shifting in and outside of the price range of a renter is not solvable on the host side, as the price limits of a renter are usually not disclosed. When the network grows there will be many renters with their own pricing ranges.
Indexd now no longer forms contracts with hosts where the pricing is less than 20% below the pricing limits; https://github.com/SiaFoundation/indexd/pull/567

Please elaborate what these “indexd signals” would be.

Sorry, but this does not make sense at all. Are you suggesting to lower the collateral ratio if balance gets too low? If so, it is advised against lowering below x2: Discord

If by “rebalancing” you mean adding extra funds, a Siagraph alert would have the same effect.

In the discussion where my own quote from April 10th came from, I explained that once egress usage kicks off hosts will be less likely to have their funds starved from contract growth.

Please explain what information Indexd discloses about network state and corresponding host health.

Siagraph informs you within 1-2 hours if the current balance could result in any active contract not being renewed. This is far ahead of any potential renewal failure.

Automatic rollback is just asking for trouble. And as stated earlier, proper migrations would cover this entirely.

All three quotes are bug related and not something a user or AI agent could have prevented.

Please elaborate.
I only focused on the host section the last time, but I can’t make any sense of how the agent relates to anything Indexd.

With all due respects, it’s fine to use LLM to help you write a text, but throwing feedback and a year of Discord chat logs into an LLM won’t be able to replicate actual hosting experience.

Please follow Mike’s suggestion to get some hosting experience yourself before you continue to try to improve that experience.