Summoner

Executive Summary

Summoner is building coordination infrastructure for the "AI workforce" — a decentralized protocol (SPLT) that enables AI agents to discover, negotiate, and transact across organizational boundaries with durable state and cryptographic trust. While most agent frameworks focus on making individual agents smarter, Summoner focuses on the coordination layer between them. ^[1]

The philosophy is explicitly "state over prediction" — decisions should be executable, inspectable, and enforceable, not probabilistic. ^[2]

Attribute	Value
Company	Summoner Network
Founded	~2025
Funding	Not disclosed
License	Apache 2.0
GitHub Stars	~50 (across repos)

---

Product Overview

Summoner's core offering is the SPLT Protocol — a coordination protocol that sits above MCP (tool access) and A2A (agent-to-agent messaging) to handle cross-trust-boundary coordination with durable state. ^[3]

The key architectural bet: agents run on user-controlled nodes (edge/local), connected via untrusted relay servers. The relay handles network physics (connections, routing) but never sees inside encrypted payloads. Validation happens at the edge, in the agents themselves.

Key Capabilities

Capability	Description
Signed Decision Graph	Every state transition is cryptographically signed and stored — not fire-and-forget
Self-Sovereign Identity	Agents generate Ed25519 keys; no centralized identity authority
Reputation-Aware Discovery	Agents discover peers based on verified history and capabilities
Dual Runtime	Python (asyncio) for dev, Rust (Tokio) for high-concurrency production
Adaptive Backpressure	Three-stage defense (Throttle, Flow Control, Disconnect) for fleet stability
Privacy-First Observability	Semantic log filtering — route metadata without payload exposure

Architecture

Agent Fleets (Local)          Decision Network (Cross-Org)
┌─────────────┐              ┌─────────────────────────┐
│ Your agents  │◄──────────►│  Relay Infrastructure    │
│ (SAP, WMS,   │  SPLT       │  (untrusted, encrypted)  │
│  Slack, etc.) │  Protocol   │                          │
└─────────────┘              └─────────────────────────┘

Product Surfaces

Surface	Description	Status
Python SDK	Core agent development library	Available
Rust SDK	High-concurrency production runtime	Available
Desktop App	Visual agent fleet management	Early stage
Relay Infrastructure	Encrypted agent routing network	Available

---

Technical Architecture

The SPLT protocol operates on a decentralized mesh model: ^[3]

1. Agents run locally on user-controlled nodes, connecting to existing enterprise systems
2. Relays are untrusted servers that route encrypted messages between agents
3. Decision Graphs capture durable, signed state — long-lived sessions that persist across connection drops
4. Discovery is reputation-aware — agents find peers based on verified history, not static registries

Key Technical Details

Aspect	Detail
Protocol	SPLT (Summoner Protocol for Long-lived Transactions)
Identity	Self-sovereign, Ed25519 key pairs
State	Signed decision graphs (cryptographically verifiable)
Runtimes	Python (asyncio) + Rust (Tokio)
Compatibility	LangChain, CrewAI, AutoGen, raw LLM calls
Open Source	Yes (Apache 2.0)

Code Example

@agent.receive("ready --> action")
async def handle_action(ctx):
    if ctx["status"] == "done":
        return Move(Trigger.done)

@agent.send("ready --> action", on_actions={Action.MOVE})
async def send_msg():
    return {"qty": order.quantity, "meta_data": {"ts": "01/01/2026"}}

---

Strengths

• Novel positioning — Coordination-as-infrastructure is an underserved layer; most frameworks focus on agent intelligence, not inter-agent trust
• Cross-organizational — Designed for agents that span company boundaries, not just internal teams
• Durable state — Signed decision graphs survive connection drops, unlike ephemeral A2A/MCP sessions
• Self-sovereign identity — Ed25519 keys avoid centralized identity bottlenecks
• Framework-agnostic — Works with LangChain, CrewAI, AutoGen, or custom agents
• Dual runtime — Python for rapid dev, Rust for production performance
• Privacy-first — Relay servers are explicitly untrusted; end-to-end encryption by design

---

Cautions

• Extremely early — ~50 total GitHub stars across all repos; no disclosed users or deployments
• No disclosed funding — Unknown financial runway or backing
• Enterprise claims unproven — Homepage mentions enterprise operations but no customers named
• Complexity — Signed decision graphs, SPLT protocol, Ed25519 identity — significant learning curve
• Sparse documentation — Limited real-world examples and tutorials
• Protocol adoption risk — Competing against established standards (A2A, MCP) with network effects
• Desktop app stale — summoner-desktop last pushed Nov 2025
• Small team — Organization activity suggests very small team

---

Pricing & Licensing

Tier	Price	Includes
Open Source	Free	Core SDK, relay, agents (Apache 2.0)
Enterprise	Unknown	Not yet announced

Hidden costs: Self-hosted relay infrastructure; compute for agent fleets

---

Competitive Positioning

vs. Adjacent Protocols

Protocol	Layer	Summoner Differentiator
MCP	Tool access	Summoner sits above MCP — coordinates agents that use MCP tools
A2A (Google)	Agent-to-agent	A2A is stateless/ephemeral; SPLT provides durable signed state across sessions
ACP	Agent conversations	ACP focuses on conversations; SPLT focuses on cross-org transactions with trust

vs. Agent Frameworks

Competitor	Differentiation
CrewAI	CrewAI orchestrates agents within a single application; Summoner coordinates across organizational boundaries
AutoGen	AutoGen is a multi-agent framework; Summoner is a coordination protocol that AutoGen agents could use
LangGraph	LangGraph manages agent state locally; Summoner manages distributed state across trust boundaries

When to Choose Summoner

• Choose Summoner when: You need agents from different organizations to coordinate with durable trust and state
• Choose A2A when: You need standard agent-to-agent messaging within a trusted environment
• Choose CrewAI/AutoGen when: You need multi-agent orchestration within a single application
• Choose MCP when: You need agents to access tools and data sources

---

Ideal Customer Profile

Best fit:

• Enterprises running agent fleets across multiple systems (SAP, WMS, Slack)
• Organizations needing cross-company agent coordination (supply chain, logistics)
• Teams building multi-party agent workflows requiring audit trails and signed state
• Developers who want to avoid centralized agent hosting platforms

Poor fit:

• Teams building single-agent applications
• Developers wanting simple agent frameworks without distributed systems complexity
• Organizations needing production-ready, battle-tested infrastructure today
• Teams without infrastructure expertise to run relay networks

---

Viability Assessment

Factor	Assessment
Financial Health	Unknown — no disclosed funding or revenue
Market Position	Pioneering — "coordination infrastructure" is a nascent category
Innovation Pace	Active — repos updated weekly, protocol evolving
Community/Ecosystem	Minimal — ~50 stars, no visible community
Long-term Outlook	High-risk/high-reward — correct thesis but very early execution

Summoner is betting that coordination, not intelligence, is the bottleneck for enterprise AI agent deployment. This is a defensible thesis — as agent count scales, orchestration becomes the hard problem. But the execution is extremely early, the team is small, and competing against Google's A2A with network effects is a steep climb.

---

Bottom Line

Summoner tackles a genuine gap: cross-organizational agent coordination with durable trust. The SPLT protocol's signed decision graphs and self-sovereign identity are technically interesting and differentiated from A2A/MCP's simpler models.

Recommended for: Researchers and early adopters exploring cross-organizational agent coordination protocols.

Not recommended for: Anyone needing production-ready agent infrastructure today.

Outlook: The coordination layer thesis is sound — enterprise agent deployments will need this. But Summoner needs funding, users, and ecosystem adoption to survive against A2A's Google backing and the broader agent framework landscape. Watch for enterprise partnerships or protocol adoption by major frameworks as validation signals.

---

Research by Ry Walker Research • methodology