Where Stallari is going

The dispatch core: durable, replay-safe, accountable. Swift harness and vault engine over a dispatch engine built on hash-chained receipts, compensating sagas, and risk-gated judgement — with a full trace store underneath.

• ✓ Hash-chained work receipts (tracks) — every step auditable, never silently rewritten
• ✓ Exactly-once dispatch — a job that fires twice still lands once
• ✓ Compensating sagas — a failed step unwinds the completed ones in reverse; a failed compensation surfaces loudly, never silently
• ✓ Risk-gated judgement — every consequential tool call passes a gate; the highest-risk actions wait for you
• ✓ Trace store — full tool-call history, token counts, and cost per run

Memory and search: what your agents learn compounds, and what you’ve written is always findable.

• ✓ Agent memory with importance decay — what matters strengthens, what doesn’t fades
• ✓ Recall and consolidation — memories resurface in context and merge as patterns emerge
• ✓ Hybrid search across your corpus — keyword and semantic together; find anything, with context
• ✓ Post-dispatch memory pipeline — every run leaves the system a little smarter

macOS app — published, notarised, Developer ID signed. Scheduled and on-demand dispatch, digest views, setup wizard, fleet sidebar, plugin management, and privacy controls.

• ✓ Agents run on a schedule or on demand — mornings start with the work already triaged
• ✓ Digest views — every scheduled run summarised; drafts wait for your review
• ✓ Privacy controls with content redaction and domain-scoped exclusions
• ✓ Secure Enclave credential store for providers and tools

Run Stallari across macOS devices in a fabric with automatic leader election and health monitoring.

• ✓ Discover peers automatically via Tailscale or Bonjour
• ✓ Enrol devices with QR scan and optional 2FA — per-peer tokens, individually revocable
• ✓ Three-tier leader election with automatic dispatch failover
• ✓ Fabric health monitoring with anomaly detection

Plugin marketplace, commerce, identity, and distribution. 20+ MCPs available — including community contributions for Home Assistant, Shopify, Xero, Gmail, iMessage, Cloudflare, and more.

• ✓ Marketplace with conformance tiers — MCP tools and Packs, portable agent workflows that reference plugins by contract
• ✓ Organisation packs — teams share private packs scoped to their org (built; unlocks fully when federation lands)
• ✓ Marketplace sign-in (GitHub and Apple) with Paddle checkout and licensing — community pack creators are paid directly
• ✓ Developer ID signed and notarised, with in-app self-updates
• ✓ Opt-in crash reporting with PII scrubbing, health probes, and process guardian

Zero-config secure networking between your tools and devices, on your choice of mesh. Automatic HTTPS, cryptographic request signing, and cross-machine trace sync.

• ✓ Cryptographically signed service-to-service requests
• ✓ Reliable trace sync across machines
• ✓ Fabric transport layer — Tailscale, Nebula, or NetBird, your choice of mesh
• ✓ Cloudflare Tunnel — expose services without a traditional VPN
• ✓ Pulse service mesh — the live channel your devices and agents use to find and talk to each other in real time

Your agents, your providers, your rules. Mix frontier providers — Anthropic, xAI, OpenAI, Google — with local models inside a single workflow. Pack-driven orchestration means workflows are portable YAML, not vendor lock-in. Scope-based routing ensures the right agent handles the right work.

• ✓ On-device inference with MLX on Apple Silicon (Qwen3.5, Phi-4, Mistral Nemo) — local models as a first-class provider, with LM Studio for power users
• ✓ Pack-driven orchestration — define agent workflows as portable YAML manifests
• ✓ Chat interface with conversation persistence and multi-provider backends
• ✓ Visible meter — every dispatch counts its tokens and its cents
• ✓ Signal primitives in the pack spec — packs can already declare the events worth your attention, and community packs can use them today
• → Webhook-triggered dispatch — packs declare the inbound events they handle (GitHub, Xero, Paddle, Square, Shopify, and more) and auto-wire them into workflows
• → Distributed inference across your fabric — dispatch a model call to whichever Mac has the headroom

Your intelligence is yours. Full-fidelity export of every memory, association, and learning your agents have built — in open formats, with documented schemas, importable anywhere.

• ✓ Signed, portable export — if you leave, your intelligence leaves with you, today
• → One-command memory export — Markdown, JSON, or SQLite, with full association graph
• → Documented memory schema — published spec so other tools can read your memories
• → Memory import — bring learnings from other systems into Stallari
• → Association graph visualisation — explore what your agents have learned and how concepts connect

Extend beyond your own fabric — and beyond the desk. Organisation members expose selected services to each other, your Chief of Staff follows you off the Mac, and the noise gets triaged before it reaches you.

• → iPhone interface — the Chief of Staff in your pocket, seamlessly drawing on inference served by the Macs in your fabric
• → Multi-vault support — keep work, home, and each client in separate vaults, each with its own corpus, memory, and rules
• → Inter-user coordination — scoped, bound agent-to-agent protocols; reach that extends on your terms
• → Signals triage — building on the signal primitives already in the pack spec, a layer that watches your tools and events and raises only what genuinely needs you