How Digital Partner's local, sovereign, verifiable memory answers the real needs of professionals.
You run a local LLM (llama.cpp, Ollama, Phi-3.5-mini) for professional work. You value its speed, privacy, and ability to process sensitive data without the cloud. But the LLM has limited memory: it can only keep a fraction of your exchanges in context (4k to 128k tokens depending on the model). Once the conversation ends, the context is gone. You cannot retrieve something discussed days ago, nor justify a past decision with the sources that motivated it. In a professional setting, this amnesia is an operational and legal risk: a law firm cannot prove that the assistant actually considered a precedent cited three months ago; an R&D lab cannot recover why a hypothesis was dropped; a public service cannot reconstruct the decision chain behind an answer.
Digital Partner brings persistent, governed, provable memory to your local LLM. Persistent: every exchange is kept in durable associative memory, reachable by semantic similarity (HNSW+PQ) or exact query. Governed: you define retention policies (duration, segmentation, purge) in a YAML file executed by the Tosio engine. Provable: every exchange is timestamped and cryptographically signed (SHA-256), creating an unforgeable proof chain.
Digital Partner remembers client cases, cited precedents (Judilibre: 480,000 decisions, Legifrance, LibreJustice), and arguments made. Every consultation is traced and timestamped.
Experiment results, tested protocols, discarded hypotheses: everything is stored and findable by semantic similarity. Digital Partner alerts you when a new hypothesis contradicts an earlier observation.
Digital Partner indexes care protocols (HAS, Guideline Central, ECRI Guidelines Trust, Clinical Guidelines Corpus: 46,000+ directives) and cross-references them with the patient context. Every interaction is traced with a SHA-256 proof, respecting medical confidentiality.
Mainstream AI assistants — ChatGPT, Copilot, Gemini — centralize your data on remote servers. You lose control of your history, your preferences, and the confidentiality of your exchanges. For a professional handling sensitive data — lawyer, physician, executive, researcher — this outsourcing is unacceptable, and in some regulatory contexts illegal (GDPR, medical confidentiality, professional secrecy). Worse: these assistants don't know your personal context — they answer generically, with no memory of your files, your response preferences, or your business constraints. You spend more time restating context than getting useful answers.
Digital Partner becomes your sovereign personal assistant. It runs exclusively locally, knows your professional context, respects your confidentiality rules, and even works offline. You define three operating modes: private (strictly local — no data leaves), hybrid (local + cloud API for selected non-sensitive tasks, with explicit consent), or full local with your own LLM. The cognitive router automatically picks the processing path: 60% of requests are resolved in < 10 ms with no LLM call at all.
Digital Partner indexes care protocols (HAS, Guideline Central, ECRI Guidelines Trust, Clinical Guidelines Corpus: 46,000+ clinical directives) and drug interactions, without ever sending health data to a cloud. HDS compliance ensured by 100% local processing.
Digital Partner indexes case law (Judilibre: 480,000 decisions since 1947, Legifrance, LibreJustice open source, OpenLegi: 1.2M decisions) and cross-references it with client files. Every search is timestamped and provable, without ever exposing defense strategies.
Digital Partner indexes your reading, notes, and hypotheses. Through the associative W dimension, it alerts you when a new publication contradicts or confirms an earlier hypothesis.
Experts — consultants, project managers, engineers, architects — juggle dozens of files simultaneously. Information is scattered across emails, notes, meeting minutes, PDFs, Slack messages, and informal conversations. When a client calls with a specific question about a decision made three months ago, finding the exact context is a nightmare. Worse: during handover or audit, reconstructing the decision timeline is nearly impossible — emails are in the former employee's mailbox, notes on their hard drive, and informal conversations lost. The cost of this chaos is concrete: time wasted searching, errors from incomplete information, and legal risk when commitments cannot be traced.
Digital Partner centralizes the complete context of each project in an isolated SafeSwarm. Automatic document extraction (PDF, Word, emails, images with OCR), indexing by the HNSW+PQ engine (4× compression, 1 million vectors = 61 MB instead of 244 MB), and full decision traceability via the ProofArchive. Each project has its own YAML governance policies executed by Tosio: who can read, modify, export, and under what conditions (quorum, thresholds, temporary delegation).
Each engagement has its own isolated SafeSwarm. Digital Partner instantly retrieves discarded hypotheses, figures, and validated recommendations — with date, author, and source document. Strict segmentation guarantees no information leaks between competing clients.
Architectural decisions are automatically indexed. Digital Partner explains why a technical choice was made 6 months ago, with the context of the time and the alternatives considered — every decision is timestamped and signed.
Plans, quotes, site reports, progress photos: everything is indexed and queryable. Digital Partner retrieves the version of the plan discussed on April 12, with the annotations and decisions made that day — the SHA-256 proof guarantees document integrity.
In regulated sectors — finance, healthcare, legal, defense, critical industries — compliance is not optional. Every decision must be justifiable, every access to sensitive data must be traced, every modification must leave an indelible footprint. Yet current tools produce flat logs, easy to alter, and disconnected from business context. In an audit, reconstructing the full history of a decision takes days, even weeks. Regulators (ACPR, AMF, CNIL, HDS, ITAR/EAR) now demand proactive traceability — the burden of proof lies with the organization, not the auditor. Without a native proof system, you are in permanent risk of non-compliance.
Digital Partner ships native, local, tamper-proof audit capability. Every action — consultation, modification, decision — is recorded in an immutable SHA-256 hash-chained journal. Governance is defined in executable YAML: quorum, thresholds, temporary delegation, access conditions. The Tosio engine checks every action before execution — no after-the-fact cleanups. Audit export is instant and includes the complete causal chain, recomputable from the export without system access. Optional external anchoring (blockchain, qualified eIDAS timestamping) strengthens evidentiary value for the most demanding use cases.
Validating a risky transaction: Tosio checks that the required quorum (2 of 3 signatories) is met, that amount thresholds are respected, and logs the decision with SHA-256 proofs. During an ACPR/AMF inspection, the audit report is generated in seconds.
Access to patient records: the Access Gate verifies the legitimacy of access (attending physician, emergency) and consent. Every consultation is traced with timestamp and SHA-256 proof. GDPR compliance and HDS certification made easier.
Software build decision: Tosio enforces security policies (DevSecOps quorum), verifies classifications, and logs every pipeline step. ITAR/EAR compliance audits are automated — every decision is reconstructible.
Software vendors and SaaS platforms are embedding more and more AI in their products, but governance of that AI remains a blind spot. Who is responsible for an AI-made decision? How do you prove a model wasn't biased by contaminated data? How do you offer your clients full traceability without building an entire audit infrastructure? Regulators (EU AI Act, GDPR, sector directives) now demand traceable governance — and vendors must comply without slowing their time-to-market. Building this governance layer from scratch means months of development, with high compliance risk if implemented badly.
Digital Partner offers an embedded governance SDK. Add a layer of traceability, proof, and access control to your own product in a few lines of code. The SDK exposes Digital Partner's native components — GovernanceBlock (block/quarantine in 0.34 ms), ProofArchive (SHA-256 hash-chained journal), Access Gate (three-level hardened access portal) — through a simple Python API. Governance policies are defined in portable YAML — your clients can customize them without modifying your code. The proof journal is embedded in your application, and audit exports are standardized.
Your HR platform includes an AI assistant for employee questions. With the Digital Partner SDK, every AI answer is traced with a SHA-256 proof. Your clients (HR directors) can audit the answers and prove GDPR compliance — without you having to build that infrastructure.
Your clients build business workflows on your platform. The Digital Partner SDK adds native governance: quorum validation, temporary delegation, approval thresholds. Every workflow is auditable effortlessly — the GovernanceBlock checks in 0.34 ms.
Your diagnostic support software uses an LLM to suggest leads. The Digital Partner SDK traces every suggestion with the clinical context, the model used, and the physician's final decision. Easier compliance with Medical Device and HDS regulations.
In industrial, IoT, and robotic environments, decisions must be made locally, quickly, and with minimal energy consumption. But how do you prove that a decision made by an automated system or robot was the right one, especially when network connectivity is intermittent? How do you audit a fleet of 10,000 sensors without a permanent connection? How do you reconstruct the decision chain of equipment that ran offline for weeks? Cloud solutions are unsuitable: unacceptable network latency for real-time, connectivity dependency, and the energy cost of data transfer. Existing local solutions lack traceability and governance.
Digital Partner includes an ultra-lightweight edge mode for local decision-making: minimal memory (footprint < 50 MB RAM, single-core CPU is enough, no GPU), unique cryptographic identity per device (Ed25519), and a local journal with SHA-256 proofs. Decisions are made locally by the Tosio engine applying pre-loaded YAML policies — zero network latency. The local ProofArchive keeps working even in complete isolation. As soon as connectivity returns, deferred synchronization sends the full journal to the central node via the Distributed Bus V4. Cognitive routing resolves 70.7% of requests locally, without an LLM call (5 seeds, `scripts/bench_e2e_soliton_real.py`) — measured on routing, not on energy consumption.
A vibration sensor on a turbine detects an anomaly. Digital Partner Edge decides locally to trigger a maintenance alert, logs the decision with sensor data and a SHA-256 proof. Even if the plant network is saturated, the decision is traced — and resolving 70.7% of requests locally (without an LLM call) is critical on a battery-powered sensor.
A logistics robot makes real-time navigation decisions. Every decision (obstacle avoidance, route change) is logged locally with a SHA-256 proof. In case of incident, the complete timeline is reconstructible — even if the robot was offline.
Moisture sensors in a vineyard trigger irrigation. Decisions are made locally (no cloud in the fields), logged, and synchronized each night through a LoRaWAN gateway. The farmer can audit every trigger and prove compliance with environmental standards.
Your organization holds thousands of documents — reports, studies, standards, contracts, patents, technical documentation — but they sit dormant in network folders, SharePoint, and shared drives. Finding them takes time, comparing them is tedious, and nobody knows exactly what exists. When you prepare a commercial offer or a tender response, you lose hours searching for “that report we did for client X in 2023.” Classic search engines (Windows Search, Google Drive) don't understand document meaning — they match keywords, not concepts. Result: you miss critical information because the document uses different terminology than your query.
Digital Partner turns your document corpus into active memory. Automatic multi-format indexing, cross-document semantic search (HNSW+PQ), automatic summarization, version comparison, and extraction of constraints and commitments. Your document base becomes queryable in natural language, with sourced, timestamped answers. Unlike a classic search engine, Digital Partner understands meaning — a search for “banking cybersecurity” also finds documents about “DORA compliance” or “operational resilience.” The Zero-LLM cognitive router handles 60% of document requests in < 10 ms without calling the LLM.
Before a commercial proposal, Digital Partner finds all similar past engagements, rates charged, typical deliverables, and lessons learned. The proposal is drafted in 30 minutes instead of 3 days — with SHA-256 proofs for every cited reference.
Searching for contract clauses: “Find all our contracts that contain a termination-without-notice clause.” Digital Partner finds the exact clauses via hybrid search (semantic + symbolic), compares them, and flags differences in wording.
Standards comparison: “What are the differences between standard NF EN 1992-1-1 (2015) and the 2023 version?” Digital Partner extracts and juxtaposes the modified sections, highlighting changes — every modification is traced.
Traditional CRMs are resource-hungry: tedious manual entry, required fields, rigid workflows. As a result, they are poorly filled, or not filled at all. The critical information — what the client actually said, what was negotiated, the promises made — stays in emails, meeting minutes, or worse, in salespeople's heads. When a salesperson leaves, the client knowledge leaves with them. The cost of that loss is massive: months of onboarding for the replacement, errors from incomplete information, and frustrated clients having to explain everything again. Existing CRMs capture structured data (revenue, pipeline) but fail to capture the richness of the informal exchanges that build client relationships.
Digital Partner becomes your natural CRM. It automatically indexes all your exchanges with each client — emails, minutes, proposals, contracts — and organizes them by client/project in isolated SafeSwarms. No more manual entry: Digital Partner knows your clients, the history of commitments, past offers, and decisions made. Segmentation is strict: each client has their own SafeSwarm, sealed off from the others — it's impossible for information about client A to appear in a response about client B. Ask it in natural language, as you would your best salesperson. The ProofArchive guarantees that every promise, every deadline, every condition is timestamped and provable.
Digital Partner knows each client: validated budgets, creative approvals, past feedback. Before a client call, the account manager asks: “Summarize the last 6 months with client X.” — full brief in 10 seconds, every piece of information sourced.
Each client file is an isolated SafeSwarm. Digital Partner retrieves exchanges about the 2024 tax return, follow-ups, and documents provided. Strict separation guaranteed by YAML policies: no information leakage between clients.
Engagement tracking: Digital Partner knows contracts, amendments, and steering committee minutes. “What are the watch points for our 3 largest clients?” — Digital Partner extracts weak signals from recent exchanges via semantic analysis.
IT departments are flooded with level-1 requests: “How do I reset my password?”, “The VPN doesn't work”, “I can't access the network share.” These repetitive questions represent 60 to 80% of IT ticket volume and consume precious time that teams could spend on higher-value work. Existing SaaS chatbots (Zendesk, Intercom, Drift) outsource your data to the cloud, which is unacceptable for sensitive IT data (passwords, network configurations, internal procedures). On top of that, these chatbots are generic: they don't understand your specific context, your internal procedures, or your organization's jargon.
Tosio Chat Service is a sovereign, 100% local ITSM chatbot powered by your own knowledge base. Three-pillar architecture: Tosio RAG Engine — hybrid vector (HNSW+PQ) and symbolic retrieval with timestamped SHA-256 proofs for every answer; Tosio Swarm Memory — visitor session persistence with anonymous profiling and cross-session correlation via the associative W dimension; local LLM (llama3.2, Phi-3.5-mini via Ollama) for contextual generation. Zero personal data collected, zero cloud, 100% local processing. The FAQ grows automatically: frequent questions are promoted to dynamic FAQ entries.
3,000 employees, daily IT questions. Tosio Chat Service answers 80% of level-1 requests without human intervention — thanks to the Zero-LLM cognitive router handling 60% of requests in < 10 ms. Employees get an immediate answer, 24/7, with a timestamped proof.
Your clients ask questions about your software. The chatbot indexes your technical documentation and FAQs. Precise, sourced answers with links to the relevant documentation pages. No client data passes through the cloud.
Students ask about enrollment, exams, scholarships. Tosio Chat Service answers in real time, relieves the human helpdesk, and generates a dynamic FAQ that evolves with student questions — without collecting personal data.
Multi-site organizations, distributed teams, and edge infrastructures need to synchronize data and decisions without depending on a central cloud. But building a reliable distributed infrastructure is complex: peer discovery, distributed consensus, communication encryption, fault tolerance, node reputation. Most existing solutions (centralized VPNs, private clouds, SD-WAN) impose a central server or certification authority — a single point of failure and control. If that central point goes down, all communication stops. If it is compromised, all security collapses.
Digital Partner includes a native P2P network layer with no central authority. Self-discovering topology: nodes find each other without manual configuration via mDNS (LAN) or bootstrap node (WAN). Distributed Ed25519 consensus: every node signs its decisions, quorum (66%) is verified without a leader. Encrypted VPN tunnels are native for all inter-node communications. Node reputation is computed automatically (response time, availability, signature accuracy). ZAN (Zero-trust Autonomous Network) handles self-repair: low-reputation nodes are progressively isolated, traffic is rerouted, and the topology reconfigures automatically. SwarmRouterV4 supports up to 5 simultaneous failures across 8 nodes without routing interruption.
A regional bank with 50 branches. Each branch runs a Digital Partner node. Credit decisions are validated by quorum (branch manager + risk manager + compliance) with SHA-256 proofs and P2P distribution between machines. No data passes through a central cloud.
Humanitarian teams in areas with intermittent connectivity. Each team runs a Digital Partner node on a ruggedized laptop. Decisions are made locally, then transferred manually when the satellite connection is available (automatic synchronization is not implemented as of today) — the local ProofArchive ensures continuity.
5 factories in 3 countries. Each factory runs a Digital Partner node for predictive maintenance. Sensor data is shared between machines through P2P distribution (registration, trusted peers). Inter-factory consensus validates production stop decisions — without depending on a central cloud, with tolerance for 5 simultaneous failures.
In decentralized finance (DeFi) and regulated markets, front-running and MEV (Miner Extractable Value) are critical problems. A malicious actor can observe a pending transaction and place another one just before it to profit. Regulators (MiCA in Europe, SEC in the US) now demand complete traceability and protection against these manipulations. But how do you prove the exact order of transactions without relying on a centralized third party? Existing solutions (private relays, encrypted mempools) add complexity and create new trust points. The fundamental problem remains: transaction timestamping is fallible because it depends on a manipulable external clock.
Digital Partner provides a cryptographic trust building block with CTD (Deterministic Time Counter), generating HMAC-SHA256 proofs timestamped by time pulse. Unlike a simple timestamp, the CTD is self-sufficient: it derives its pulses from the local system clock and the internal HMAC chain, without depending on an external time server. Each transaction receives a unique, unforgeable time nonce linked to the HMAC chain. Clock rollback detection is automatic: if the system clock is manipulated (rollback), the CTD detects the inconsistency. The causal chain is verifiable offline: the exact transaction order can be reconstructed from the HMAC-SHA256 proofs, without access to the original system. Anti-replay is native: a transaction with an already-used nonce is automatically rejected.
Each order receives a CTD time nonce before being sent to the order book. In a dispute about order priority, the HMAC-SHA256 proof is irrefutable. MiCA compliance without external infrastructure.
Transactions submitted to a private mempool receive a CTD nonce. Even if a miner tries to reorder transactions, the causal order is verifiable via the HMAC-SHA256 chain. Manipulation is cryptographically detectable.
Clearing orders between banks receive timestamped CTD nonces. The complete causal chain is auditable by the central bank or regulator, without access to the banks' internal systems.
Business processes are modeled in BPMN 2.0, but their execution is fragmented: some run on Camunda, others on Temporal, others in ad hoc Python scripts. Each engine manages permissions, traceability, and compliance differently. Result: no unified view of process compliance, and every engine change means rebuilding everything. Worse: how do you prove that a critical process (credit validation, drug batch release) ran in compliance with the rules? Each engine's logs are incompatible, and none provide cryptographic proof.
Digital Partner becomes the certified conductor of your business processes. It parses and validates BPMN 2.0 locally — role checks, scopes, classifications, and SHA-256 proof generation — then delegates execution to the engine of your choice via a normalized JSON contract (Tosio format). Supported engines: SpiffWorkflow (local, free, embedded), Camunda, Temporal, n8n, or your own API. No matter the execution engine: Digital Partner guarantees compliance (policy checks before every step), traceability (timestamped SHA-256 proofs at every transition), and auditability (complete compliance report at process end). Change engines without changing governance.
BPMN workflow modeled with 5 steps: intake, risk analysis, committee, management validation, disbursement. Digital Partner validates roles and thresholds before each step. Execution delegated to Camunda. Every decision is timestamped with a SHA-256 proof. Easier ACPR audits.
Drug batch release process modeled in BPMN with quality checks, regulatory verification, and electronic signature. Digital Partner guarantees GMP (Good Manufacturing Practices) compliance and logs every step — irrefutable proof in case of inspection.
Simple BPMN workflow: KYC verification, account creation, welcome email, account manager assignment. Digital Partner executes locally via SpiffWorkflow (free, embedded). GDPR compliance proofs generated automatically at each step.
Freelancing platforms, B2B marketplaces, and professional forums all suffer from the same issue: reputation systems are biased, manipulable, and unreliable. Paid fake reviews, mutual back-scratching between friends, retaliation after a negative review — nothing guarantees that a rating reflects the real quality of a service. Worse: on most platforms, anyone can leave a review, even without ever interacting with the provider. Trust in these systems is near zero, which penalizes excellent providers who don't (yet) have a network to generate reviews.
Digital Partner acts as an infrastructural trust third party for rating. It certifies transactions (QLoRA jobs, API calls, service engagements) with a SHA-256 proof, then aggregates mutual evaluations between parties that actually interacted. You cannot rate without participating in a certified transaction — every evaluation is cryptographically linked to a specific transaction. Scores are verified, anonymous, and unforgeable: both parties rate each other without knowing the score the other gave. Sector benchmarks are generated automatically from real transactions, not reviews. Badges (Verified client, 10+ jobs, Sector expert) are awarded automatically by Digital Partner, not self-declared.
A client submits a QLoRA adaptation need. A provider completes the job. Both parties rate each other after transaction certification. The provider's score reflects the real quality of their adaptations — not reviews from friends.
Consultants offer their services. Each engagement is certified by Digital Partner (contract, execution, deliverables). Ratings are computed on real engagements. Badges (10+ engagements, Verified client) are awarded automatically — unforgeable.
API providers expose their services. Every call is certified by Digital Partner. Consumers rate reliability, latency, and documentation. Scores are public and verifiable. Bad APIs are naturally pushed out by the market.
Creators of new media formats — interactive podcasts, augmented newspapers, branching documentaries, citizen debate platforms — want to offer their audience a rich experience without depending on complex cloud infrastructure. To do that, they need advanced features: preference memory, contribution proof, editorial decision traceability, access governance. Building these blocks from scratch takes months and costs a lot — especially for lightweight organizations (independent media, solo creators, associations).
Digital Partner exposes an open, documented REST API that provides these key components out of the box. Memory: store and retrieve your audience's interactions via the Swarm API. Proof: every contribution is SHA-256 timestamped — ideal for public debate or voting. Governance: define your access rules in YAML via the GovernanceBlock API (0.34 ms check). Everything runs locally — no cloud, no recurring fees, no call limits. Your application controls everything through simple HTTP calls.
http://localhost:8000.POST /analyse to store, GET /audit to audit. Swagger documentation at /docs.An investigative podcast invites listeners to contribute to the investigation by submitting documents or testimonies. Every contribution is SHA-256 timestamped via the API, publicly verifiable. YAML governance ensures only journalists can access raw sources.
An association runs online debates on social issues. The Digital Partner API logs every argument with its context, author, and date. Moderators can audit the full history. Participants can verify their words were not modified.
An online documentary offers branching narratives. The API stores each viewer's choices and consolidates them into audience statistics — without collecting personal data, without cloud. The SHA-256 proof guarantees the integrity of viewing data.
localhost. You can use it offline, in an air-gapped environment, or behind a VPN.People who share the same interests — social innovation, participatory democracy, open governance, ethical technologies — are scattered across dozens of incompatible platforms. There's no way to know who is working on what, who solved a given problem, or who is looking for a given skill. Civic projects struggle to find contributors. Associations duplicate efforts without knowing it. Good ideas die for lack of meeting the right people.
TOtB (Think out of the Box) becomes the interest-based connection platform, backed by Digital Partner. Verified identity: every participant has a governed profile (who can see what, which skills are public). Automatic matching: Digital Partner detects complementarities between profiles through the associative W dimension — “You work on participatory democracy? Here are three people who solved similar problems.” Contribution traceability: every interaction, project, and decision is timestamped with a SHA-256 proof — reputation is verifiable, not declarative. Community governance: participants define their rules in YAML — who can join, vote, propose, decide.
A collective works on municipal participatory democracy. TOtB connects them with an association that deployed a similar system in another region — both organizations share their experience, tools, and failures.
Committed developers want to contribute to open source projects aligned with their values. TOtB matches them with projects looking for their specific skills — every contribution is traced, reputation is verifiable.
Several associations work on digital inclusion without knowing it. TOtB detects synergies and offers them a shared space — they pool resources, training, and institutional contacts.
Two operating regimes dominate last-mile logistics. Courier fleets work on a moving target: stops appear, cancel and shift all day, so every route must be reorganized continuously on a changing set of points. Warehouse robots face the opposite structure: storage locations are fixed, known and shared between all robots — 100,000 orders does not mean 100,000 points, because every unit passes through a common set of storage locations. Only the active subset of those shared locations changes with each order wave, and the real payoff comes from coordinating several robots so that routes and orders are balanced across the fleet. In both cases the same three needs return: good routes right now, instant re-planning, and auditable decisions.
Digital Partner provides one local engine for both regimes, with a clear division of roles. Tosio, the orchestrator, segments the area into islands (GeoPandas + H3 hexagonal indexing + Shapely barriers) and assigns them to units through governed YAML policies, while keeping the proof journal. Each robot or van carries its own local instance: on the fixed warehouse graph, distances and neighbours are precomputed once, and only the active-order subset is re-solved; on the variable courier graph, the fast ILS+3-opt solver rebuilds routes continuously. Between units, a traced contract-net lets robots exchange orders at the margin instead of recomputing the whole fleet. When optimality must be certified, CP-SAT proves the local island tour, with SCIP available for cross-validation. Every assignment, route and exchange is timestamped into the SHA-256 chain.
Storage points are fixed; only ~20% are active per order wave. Each robot re-solves just its active stops on the cached graph, and a fault only triggers a targeted order exchange — not a fleet-wide recompute.
The stop list changes all day (new parcels, returns, absent recipients). Each van re-plans its route live on a variable point set, with every change logged and provable.
In a larger optimization problem, a solver often needs a reliable lower bound to prune the search and prove that no better solution exists. Recomputing that bound from scratch is expensive; and a bound produced by the wrong technique is either too loose (no pruning) or unsafe (above the true optimum, which would silently hide valid solutions).
The route optimization engine already embeds a conservative DFJ/Held-Karp lower bound (`dfj_lp`/`dfj_lp_dual`) used by the SCIP-DFJ solver. It measures the true optimum from below and is safe by construction: it can never exceed it. This bound is exposed for use as an evaluator in a broader branch-and-cut, exactly like a native dual cut.
A global plan optimizes both assignment and routes; the local routing bound prunes the assignment search.
A decomposition strategy uses the bound as a quality certificate for each sub-problem.
An exact solver needs a good starting point: the better the initial solution, the faster it prunes. But building that start from nothing wastes presolve time that could be spent proving.
The engine provides ILS+3-opt as a fast heuristic upper bound, then hands it to SCIP-DFJ via `warm_ub`. CP-SAT can serve as a primal oracle for cross-validation. This is the classic “fast heuristic → upper bound → exact proof” pattern.
A good incumbent from ILS+3-opt shortens the proof for a fleet manager who must certify routes.
The same warm path calibrates quality against available time with a proof flag.
These applications fall within sensitive usage areas where we want to be involved from the start so we can support you.
| Domain | Target user | Value |
|---|---|---|
| Regulated IT teams | CIOs, CISOs, DPOs | Rights delegation, verified identity, continuous audit |
| Financial operators | Traders, risk, compliance | Validation of risky operations with thresholds and quorum |
| Healthcare and sensitive data | Physicians, hospitals | Consent, access traceability, audit |
| Fintech and payments | Ops, risk, compliance | Granular control, quorum, timestamped proof |
| Insurance | Managers, legal counsel | Consent, temporary delegation, continuous audit |
| Lawyers / Law firms | Firms, legal counsel | Case-law history, per-file rating |
| Banking / Advisory | Advisors, RM | Client context, proof of advice |
| Commercial negotiation | Sales, legal | Recording, formalization, final protocol |
| Police / Justice | Investigators, court clerks | Minutes, exchange proofs, protocols |
| Public administrations | Public services | File updates, certificates |
| Regulated supply chain | Industry, logistics | Traceability of critical interventions |
| Software compliance | DevSecOps, vendors | Security validation, build decisions |
| Automation platforms | RPA, workflows | Workflow validation with policies |
| AI tooling providers | SDK, API, LLM infra | Identity, authorization, traceability layer |
| Verifiable quantum memory | Researchers, physicists, labs | Traceable, unforgeable encoding of quantum measurements via N3+1 topological soliton and SHA-256 ProofArchive |