Spanda

Spanda Product Strategy

Strategic analysis for Spanda as an autonomous systems platform with a safety-first language at its core. This document defines positioning, priorities, and release scope. It complements vision.md (aspiration), platform-overview.md (architecture), and feature-status.md (implementation truth).

Last updated: 2026-07-04 (v0.6.3 evaluation/beta; Next horizon phase; organizational gates)


Executive summary

Spanda should not compete as a general-purpose language, Python replacement, Rust replacement, or ROS replacement. It wins as the coordination and verification platform where:

  1. AI output is typed as untrusted (ActionProposalSafeAction)
  2. Hardware fit is checked before deploy (spanda verify)
  3. Safety is mandatory in the language, not bolted on
  4. Simulation, replay, and health are first-class operational workflows

Official positioning: Spanda is an Autonomous Systems Platform with a safety-first programming language at its core. The pulse of autonomous intelligence.

Short form: The Autonomous Systems Platform.

Sharpened tagline: Build. Verify. Simulate. Deploy. Operate. — for autonomous systems.

Platform maturity focus (2026–2027)

Beyond core language features, Spanda prioritizes adoption, trust, and operations — see platform-maturity-roadmap.md, enterprise-operations-roadmap.md, and differentiation-roadmap.md. Planned areas compose existing engines (spanda-readiness, spanda-assurance, spanda-security) rather than replacing them.

Shipped (v0.6.3 evaluation/beta): release-hardening CI suite (CI Integration), honest mock-default labeling, closed code blockers — release-readiness.md. PR gate: ci-architecture.md.

NOW (Next horizon, P1): differentiation signature capabilities (mission contracts, explainability, decision audit trail); Control Center Stable promotion.

P0 (organizational, parallel): 30-day field soak + third-party security audit — organizational-gates.md · #51.

Shipped (Stable, E1–E4): Control Center (spanda control-center serve), REST v1 API, remote CLI with --version, official SDKs 0.4.2, Tauri desktop 0.6.3 (auto desktop-v* on labeled CC PRs), WebSocket/OTLP observability, compliance/digital-thread/executive reporting — see control-center.md · control-center-versioning.md · stable-hardening-enterprise-ops.md.

Deferred: VS Code Marketplace publish (needs VSCE_PAT).

Philosophy

Hardware is the body.
Sensors are the senses.
AI models are the mind.
Actuators are the muscles.
Spanda is the intelligent pulse that transforms perception, intent, and safety into action.

Spanda (Pronounced SPUN-duh (/ˈspʌndə/)) is a Sanskrit term meaning the divine pulse — the creative vibration of consciousness and energy that manifests as expansion and contraction in all entities, bridging stillness and movement within consciousness; and the first stir of awareness that creates and sustains the universe.


Signature capabilities

Six capabilities define Spanda’s market identity. Protect them in every release decision.

# Capability What it means Status
1 Safety-Typed AI ActionProposal cannot reach actuators; only SafeAction from safety.validate() can. Stable
2 Mission Contracts Missions as verifiable entities with guarantees, constraints, assumptions, invariants, objectives. Planned (NOW)
3 Readiness Engine Weighted go/no-go across health, verify, safety, trust, mission achievability. Stable
4 Continuity & Takeover Checkpoint resume, delegation, succession, safety-gated fleet handoff. Stable
5 Trust Framework Composite trust across packages, hardware, config, identity, safety integrity. Stable
6 Explainability & Audit Trail Decision → evidence → safety check → action; trace replay. Planned (NOW)

Questions Spanda uniquely answers

Question Answered by
Can this mission run? Mission Contracts + Readiness + spanda verify
Can this robot safely perform this mission? Safety-Typed AI + Safety Coverage + safety { }
Does the hardware satisfy the required capabilities? spanda verify + capability traceability
Is the system healthy enough to deploy? Readiness Engine + Health Framework
Can it run safely? Safety-Typed AI + Safety Coverage
Can it recover? Recovery Coverage + Recovery Framework
Can it continue? Continuity & Takeover
Who can take over when a robot or fleet member fails mid-mission? Continuity & Takeover + Succession Planning
Can it be trusted? Trust Framework + Certification Packs
Why should this deployment be trusted? Trust Framework + Readiness + Assurance evidence
Why did it behave this way? Explainability + Decision Audit Trail
What happened when something failed? Diagnosis Engine + Replay + Decision Audit Trail
What evidence supports deployment? Certification Packs + Assurance Cases

Full roadmap: differentiation-roadmap.md.


Five pillars (core identity)

Foundational pillars that support the signature capabilities above.

# Pillar What it means
1 Safety-Typed AI ActionProposal cannot reach actuators; only SafeAction from safety.validate() can. Enforced at compile time and runtime.
2 Hardware Compatibility Verification spanda verify checks memory, sensors, actuators, timing, battery, network, and AI model requirements against hardware profiles.
3 Simulation-First Development spanda checkspanda verifyspanda sim as the default dev loop before hardware exists.
4 Autonomous Systems Primitives robot, sensor, actuator, task, agent, safety are language keywords — not framework imports.
5 Safe Deployment Validation deploy, requires_hardware, verify { }, mission duration, and task budgets encode the ship checklist in source.

Competitive landscape

Where Spanda wins

Capability vs Python/Rust/C++ vs ROS2/Dora-RS vs Agentic AI frameworks
AI safety as types Convention only Not addressed Runtime filters only
Pre-deploy hardware verify Manual CI/scripts URDF + launch files Not addressed
Units in robot syntax Libraries Not native Not addressed
Unified design→verify→sim Fragmented toolchain Runtime graph only No hardware/deploy model

Where Spanda does not win (and should not try)

Area Incumbent Spanda stance
ML training Python/PyTorch Orchestrate via extern python
Real-time drivers C++/Rust Call via FFI; don’t rewrite
Communication at scale ROS2 Bridge, don’t replace
Physics simulation Gazebo/Isaac/Mujoco Physics-lite sim for safety testing only
Package ecosystem PyPI/crates.io/ROS Hosted index + in-repo mirror via spanda publish; 38 official packages
Native performance Rust/C++ Interpreter first; LLVM deferred

Differentiation ratings

Feature Rating Notes
ActionProposal / SafeAction Highly differentiated No mainstream language enforces this natively
spanda verify Highly differentiated Closest analog is manual checklists, not a language primitive
Physical units Somewhat unique Libraries exist elsewhere; integration with robot syntax is the value
Robot-native syntax Somewhat unique ROS2/Dora are runtime graphs, not unified typed languages
verify { } assertions Somewhat unique Between unit tests and formal methods
Built-in simulation Already common Adequate for alpha; not a Gazebo replacement
Communication model Already common ROS2 owns production transport
AI agents / goals / memory Already common Structural value; live OpenAI/Anthropic/ONNX when configured
Blockchain / provenance Already common Optional; stub only — remove from core narrative
World models Minimal runtime world_model { } + belief hook; not full knowledge graphs
LLVM / self-hosting Already common Premature for current adoption stage

Feature classification

Every major feature is classified for prioritization. See feature-status.md for implementation status.

Feature Classification
ActionProposal / SafeAction Core Identity
Safety validation (safety { }, zones, stop_if) Core Identity
Physical units Core Identity
Hardware profiles + deploy Core Identity
Compatibility verification (spanda verify) Core Identity
Autonomous primitives (robot/sensor/actuator/task) Core Identity
Battery estimation Important
Timing verification Important
Simulation Important
verify { } behavioral assertions Important
AI agents (goal, memory, skill) Important
Communication model Important
Deployment targets Core Identity
observe / fusion Important
Package ecosystem Important
Security (capabilities, signed comm) Important
Python integration Important
ROS2 integration Important
Digital twins (local replay) Nice To Have
Replay buffer Nice To Have
Provenance / audit (local) Nice To Have
C++ integration Nice To Have
Concurrency (spawn/select) Nice To Have
Generics / async / traits Nice To Have
Digital twin cloud sync Experimental (Twin Cloud SaaS + legacy upload)
Multi-agent distributed runtime Future
World models Future
LLVM production backend Future
Self-hosting compiler Future
Blockchain support Future (community packages only)

Focus tiers

Tier 1 — Primary focus (max 5)

  1. Safety-Typed AI
  2. Hardware Compatibility Verification
  3. Autonomous Systems Primitives
  4. Simulation-First Development
  5. Safe Deployment Validation

Tier 2 — Secondary

Tier 3 — Experimental (Phase 22)

Promoted from deferred to experimental with minimal runtimes and golden paths. See tier-3-experimental.md. Execution order: tier-3-priority-plan.md.

Still future: LLVM as primary path, production chain adapters, full world-model semantics, hosted Twin Cloud product GA (billing, multi-region SLA — OSS /v1/twins/* is Stable), full fleet planning, full DDS middleware, ROS replacement, complete self-hosted compiler.


Release roadmap

v1 vision

A robotics team writes a .sd program, runs spanda verify against their hardware profile, simulates it, connects one real AI provider and one real transport (Python + ROS2), and deploys coordination logic to edge hardware — with unsafe AI blocked at compile time.

v0.1.0-alpha — shipped

Must have Status
Interpreter + type checker Done
ActionProposal → SafeAction Done
spanda verify Done
Physics-lite sim Done
Showcase examples + CI Done

Optional (present): live AI/IoT when configured, experimental LLVM, hosted registry + publish mirror. Not needed: blockchain, self-hosting, fleet runtime.

v0.5 beta — target Q4 2026

Must have Status
Published VS Code extension with LSP VSIX CI + local install; only open blocker — Marketplace pending VSCE_PAT
One live AI path (OpenAI or local ONNX via Python bridge) Done — OpenAI, Anthropic, ONNX (Phase 33–35)
One documented ROS2 golden path Doneros2_golden_path.sh
spanda verify CI integration guide Doneci-verify.md
Curated killer demo (examples/showcase/killer_demo.sd) Done
Package install from remote registry (small, curated) Done — 38 packages; hosted index + spanda publish mirror
Differentiation NOW (contracts, explain, audit trail, coverage) Planned — topic guides shipped; crates not started

P1 adoption enablers (CI verify, PyO3, LSP deploy hints, showcase trim, adoption quickstart) are complete — see tier-3-priority-plan.md.

v1.0 — target 2027

Must have  
Production-quality verify on 5+ hardware profiles  
Real deploy to one edge target (Jetson or Pi)  
Python + ROS2 interop documented and tested in CI  
Stable language subset with migration guide  
Security capabilities in production comm examples  

Optional: LLVM for hot paths, twin replay export, fleet examples.
Not needed: self-hosting, full blockchain suite, Gazebo parity.


v0.5 beta — prioritized work items

Use these as GitHub issue titles. Order reflects dependency and adoption impact.

P0 — Blockers for beta credibility

# Issue Acceptance criteria
1 Publish VS Code extension to marketplace Install from marketplace; LSP diagnostics for check and verify work out of the box
2 Curate killer demo program Single .sd file: compile-time AI block + verify pass + sim with emergency stop; documented in docs/killer-demo.md
3 One live AI provider path extern python or package calls real OpenAI/ONNX when configured; mock fallback when not
4 One ROS2 golden path ros2-golden-path.md — rclpy bridge; /cmd_vel or /scan manual validation
5 Remote package registry (minimal) spanda install fetches from a hosted registry; at least 2 curated packages (spanda-openai, spanda-ros2)

P1 — Adoption enablers

# Issue Acceptance criteria
6 CI integration guide for spanda verify ci-verify.md — GitHub Actions + GitLab; --json parsed in CI
7 In-process Python FFI (PyO3) as default when built Document build flags; subprocess remains fallback — see ffi-and-ecosystem.md
8 Hardware profile picker in LSP/VS Code Deploy target hints or autocomplete for built-in profiles
9 Trim showcase to 3 flagship examples examples/showcase/README.md — safety, verify, sim
10 Adoption quickstart: wrap existing Python stack adoption-path.md — 1-sprint integration guide

P2 — Nice to have for beta

# Issue Acceptance criteria
11 Digital twin replay export (JSON) Export replay buffer for post-incident review
12 DAP debugger polish Step through task every loops in VS Code
13 Web playground: killer demo preset WASM run of killer demo without local Rust build

Explicitly out of scope for v0.5


Developer adoption

Current workflow

Python (AI, glue) + ROS2 (comm) + C++ (drivers) + manual safety monitors + deploy checklists

Spanda value

Spanda (.sd) — coordination, safety gate, verify
  ↳ extern python — PyTorch, OpenCV, existing nodes
  ↳ ROS2 bridge — existing drivers unchanged

Smallest adoption path (one sprint)

spanda check my_robot.sd
spanda verify my_robot.sd --target JetsonOrin
spanda sim my_robot.sd

Week 1: CI with check + verify.
Week 2: one extern python call to existing model.
Week 3: ROS2 bridge for a single topic.

Guides: adoption-path.md · ci-verify.md · ros2-golden-path.md

Positioning: Spanda is a 2–5k LOC coordination layer above existing stacks, not a rewrite.


Killer demo: “The Unsafe Planner”

Duration: under 5 minutes
Message: Unsafe AI is blocked at compile time; hardware fit is verified; safe execution is simulated.

Step Action Audience sees
1 Show .sd with ai_model, safety { }, deploy Readable robot program
2 spanda check (unsafe version) Compile error: ActionProposal cannot reach actuators
3 Add safety.validate(proposal) check passes
4 spanda verify --json Memory, sensors, timing, battery report
5 spanda sim Robot drives; stop_if triggers near obstacle
6 simulate_compatibility { fault ... } Verify warns before deploy

Implementation: merge examples/showcase/ai_safety_violation.sd and examples/showcase/hardware_compatibility.sd into examples/showcase/killer_demo.sd; add docs/killer-demo.md.

Do not show blockchain, world models, LLVM, or fleet comm in this demo.


Do not implement (next 6 months)

Feature Reason
Blockchain / crypto ledger packages Stub only; distracts from safety story
Self-hosting compiler No adoption value before v1
Production LLVM backend Verify + sim loop not proven in field
World models runtime No implementation exists
Digital twin cloud sync Local replay sufficient
Distributed fleet runtime Single-process examples enough
MQTT/DDS live adapters ROS2 first
Full ROS replacement Guaranteed failure vs incumbents
Advanced power models Heuristic battery check is enough
Expanding GP language surface Rust/Go win on GP; Spanda wins on domain

Exception: maintain existing LLVM/security/audit code if tests pass — but allocate no new engineering until Tier 1 is adoption-proven.


Long-term vision

Spanda becomes the standard coordination layer for autonomous systems: the .sd file between Python AI pipelines and ROS2/C++ hardware, with verify-as-CI and safety-typed AI as industry expectation.

Not a general-purpose language. Not a ROS fork. The deploy-verified, safety-gated orchestration language for robots, drones, and edge agents.