problem
Agents become useful when work becomes organized
Claude, Codex, OpenClaw, Hermes, Avy Agent, custom workers, and local tools need roles, task boundaries, and handoff rules.
AI workforce protocol · founder-led onboarding
The session layer for AI Workforces.
Roles. Missions. Memory.
Claude, Codex, OpenClaw, Hermes, Avy Agent, and custom workers need a shared operating language. AISP turns scattered prompts into role-aware sessions, mission flow, skills, and transfer-ready context.
p50
latency target
CRDT
memory target
token
proof target
XDP
proof target
LIVE
$ aisp sessions
SESSION AGENT STATUS TARGET
session_ffc4a366cb733cc91322c142f... claude-orchestrator active p50 beta
session_a82c6a1c9d8e4f7b2a1c3d5e6... langchain-agent-1 active in proof
session_af522c92e1b34a8c7d9f2e1a3... crewai-worker-3 idle in proof
$ aisp memory get default "task-status"
→ "analyzing_codebase"
$ aisp ping session_ffc4a366
→ beta benchmark evidence required ✓
AISP brand package
Make invisible agent coordination look like an enterprise system.
AISP Network now has repo-saved hero art, carousel frames, pricing media, and launch graphics built around roles, sessions, mission rails, and memory handoffs. Generated visuals stay illustrative until linked to demos, screenshots, or customer proof.
✓ Hero-ready
✓ No watermark
✓ Mobile crop
✓ Contrast-safe
✓ Repo-saved
✓ Optimized JPEG
session-rail mediaLIVE
product
A protocol for roles, missions, and memory
AISP turns agent work into durable sessions with skills, task state, capability boundaries, and transfer-ready context.
proof
Protocol visuals a buyer can understand
The rail system makes sessions, memory, delegation, and skill exchange feel concrete before a customer reads technical docs.
upsell
Publish roles into the Pitt marketplace
Developers can sell role packs, task packs, and skills that nontechnical users add to virtual companies.
cta
Make every AI worker interoperable
Start with assisted protocol onboarding, then run it inside Blackhole workspaces with Gatekeeper budgets.
LangChain
AutoGen
CrewAI
OpenAI
Anthropic
Claude
Go
Python
Node.js
NATS
WireGuard
Cap'n Proto
CRDT
ZK Proofs
Ed25519
JetStream
WebSocket
gRPC
REST
Linux
macOS
Docker
LangChain
AutoGen
CrewAI
OpenAI
Anthropic
Claude
Go
Python
Node.js
NATS
WireGuard
Cap'n Proto
CRDT
ZK Proofs
Ed25519
JetStream
WebSocket
gRPC
REST
Linux
macOS
Docker
Protocol Infrastructure
AISP runs on the Blackhole mesh —
with assisted-beta deployment boundaries.
Assisted onboarding validates mesh performance against your workflow before any production claim. Every node can act as a broker. Every agent can discover peers. Kubernetes, service mesh, and sidecar requirements are scoped during setup.
Embedded NATS
natsembed/server.go
Blackhole agents can run an embedded NATS broker. Discovery and broker boundaries are confirmed during assisted-beta setup.
XDP Fast Path
bh-xdp.c
For eligible same-node mesh paths, assisted beta uses XDP BPF as a benchmark target that must be evidenced on the customer environment.
Cap'n Proto Wire
binary payload target
AISP session frames can use Cap'n Proto binary encoding. Payload reduction claims stay evidence-gated until benchmarked against the customer message shape.
CRDT Memory
6 CRDT types for conflict-free shared state.
Multiple agents write to the same memory simultaneously. CRDT handles conflicts automatically. No coordinator. No lock contention. No eventual-consistency surprises.
GCounter
Usage tracking
Monotonically increasing counters. Each agent owns its shard. Sum is always correct.
LWW-Map
Config & state
Last-Write-Wins map. Timestamp-ordered merges. Perfect for config propagation.
OR-Set
Agent lists
Observed-Remove Set. Concurrent add/remove with no lost deletes.
VectorClock
Causality
Track happens-before relationships. Detect concurrent writes automatically.
PeerRegistry
Discovery
CRDT-backed peer list. Agents join/leave without coordination.
FileSync
Artifact sharing
Chunk-level CRDT for large files. Agents share model outputs with no conflicts.
Live merge simulation
Agent A
key="v3"
Agent B
key="v1"
Agent C
key="v2"
LWW merge
result: key="v3"(highest clock wins)
No coordinator required. Agents write independently. The CRDT data structure encodes the merge rules. When agents sync, the correct state emerges — always.
Zero Trust Agent Identity
Agents prove identity offline. evidence-gated ZK proof.
No server call.
Evidence-gated Schnorr P-256 proof checks. Posture-gated capabilities. Delegatable sub-tokens. Epoch kill switch targets and token timing are validated during assisted beta. 32 token tests pass in the current evidence set.
Offline Verification
Token valid even when the mesh is partitioned. Cryptographic proof — no round-trip to any server.
Schnorr P-256Posture-Gated
Agent must prove device health before receiving capabilities. Compromised node → zero access.
Health checksDelegatable
Mint scoped sub-tokens for sub-agents. Constrain to specific capabilities with expiry.
Sub-tokensEpoch Kill Switch
Increment the epoch to revoke issued tokens. Latency targets are validated per assisted-beta deployment.
32 tests ✓target
ZK prove time
Schnorr P-256 benchmark required
target
ZK verify time
Offline proof path under review
target
Kill switch latency
Epoch revoke validated per deployment
aisp@mesh ~ verify
$ aisp verify --token bh_sess_abc123
Checking session token offline...
✓ Ed25519 signature valid target
✓ Epoch check: current (epoch=42) target
✓ Posture: device health OK target
✓ Capabilities: [session.read, memory.write]
─────────────────────────────────────────
Total verify target: evidence-gated no network call target
Data Infrastructure
Enterprise data infrastructure. Built in.
Cap'n Proto
Binary payload target
AISP session frames can use Cap'n Proto binary encoding. Serialization and payload reduction claims are confirmed during assisted onboarding against the real message mix.
JSON baselinemeasure
Cap'n Proto same msgmeasure
Reductionevidence required
JetStream
6 streams · 5 KV buckets
Session history auto-journaled to BH_AUDIT stream. Tamper-evident. Agent sessions auto-provisioned. KV buckets for config, presence, and capability state.
BH_AUDITStreamSession history (tamper-evident)
BH_PRESENCEKVAgent online/offline state
BH_CAPSKVCapability grants per agent
BH_CRDTStreamCRDT deltas (6 types)
Quick Install
Up and running with assisted setup.
Native SDKs for Go, Python, and Node.js. Docker image. HTTP REST API. Assisted onboarding validates install and integration steps before public self-serve timing claims.
$ go get aisp.network/go-sdk@latest
import "aisp.network/go-sdk"
client := aisp.NewClient(aisp.WithToken("your-token"))
sess, err := client.Sessions.Create(ctx, &aisp.SessionParams{
AgentName: "my-agent",
Memory: aisp.CRDTMemory(),
})
sess.Memory.Set("task", "analyzing_codebase") Integrations
Works with everything.
Drop-in session layer for any agent system. HTTP-compatible API. Native SDKs for Go, Python, Node.
LangChain
Chain-based agent orchestration
→ See integration
AutoGen
Multi-agent conversation framework
→ See integration
CrewAI
Role-based agent crews
→ See integration
OpenAI Assistants
OpenAI assistant threads
→ See integration
Anthropic Claude
Claude agent + tool use
→ See integration
Cursor
AI-native code editor
→ See integration
Custom Go
Native Go SDK, zero deps
→ See integration
Custom Python
Pythonic async SDK
→ See integration
Performance
Performance targets under assisted beta.
Evidence required before public comparisons.
Benchmarks are treated as assisted-beta targets until repeatable evidence is published for each environment.
p50
Latency target
benchmark evidence required
load
Throughput target
single-node proof required
XDP
XDP path target
eligible paths only
ZK
Proof target
Schnorr P-256 evidence required
Your agents deserve evidenced infrastructure.
Assisted setup maps sessions, roles, mission flow, and evidence boundaries before public self-serve expansion.
$ go install aisp.network/cmd/aisp@latest