Altilia vs Langflow (2026)

F>P, and this is the closest call on the record, so the reasoning is recorded rather than left implicit. ONE FACT MOVED OFF THIS CELL AND IT WAS DOING MOST OF THE WORK. The July basis credited INGESTS ANY DOCUMENT OR DATA SOURCE here, listing PDFs, emails, handwritten notes, scanned images and legacy databases. That is ingestion for grounding and belongs on Know, where the Knowledge Graph carries Full. Reading a source is not integrating with a system the agent then acts in, and this record's real strength is the reading half. WHAT REMAINS IS A CATEGORY WITHOUT MEMBERS. API-first architecture and PRE-BUILT CONNECTORS are stated repeatedly, with agents EMBEDDED INTO YOUR ENTIRE TECH STACK, FROM LEGACY SYSTEMS TO MODERN CLOUD APPLICATIONS. Across two passes covering the home, platform, why-altilia, ai-agents and modules pages, not one connector is named. No CRM, ERP, ticketing or storage system appears, no catalogue exists, and no count is given. THE AXIS MEASURES BREADTH ACROSS CLASSES, and an unenumerated claim of breadth cannot evidence it. This is the same position ai-library was held at earlier today for the same reason, and consistency requires the same answer. WHAT DOES SUPPORT PARTIAL is real: the API-first architecture is a genuine integration route, agents perform actions rather than only answering, and the platform is distributed through the Microsoft commercial marketplace, which implies at least an Azure-side path. I RECORD THIS AS THE CELL MOST LIKELY TO BE UNDERSTATED ON THIS RECORD. A vendor with thirty-plus large enterprise clients in banking, insurance and public administration necessarily connects to core systems; the connector list simply is not published on the pages reached. A partner or integrations page would settle it and is the first check at lane close.

Stands at F on breadth across classes. Two structural properties make this stronger than a component count would suggest. Bundles are organised BY SERVICE PROVIDER, so integration breadth grows by provider rather than by hand-built connector, and the MCP client means any MCP-exposed tool becomes callable without Langflow shipping anything. The escape hatches are unusually open even by open-source standards: an API Request component for any endpoint, and full custom Python components for anything else, with the vendor explicitly suggesting a legacy component's code as the starting point for your own. Tool Mode on a component converts it into an agent-callable tool, which is the tool-calling half of this axis and is a clean design. Recorded per section 7: the Apify and similar bundles are Langflow reaching out to third-party services, which counts here and is not evidence of Langflow's own extensibility, graded on Ext.

Stands at F, and the orchestration primitive is named rather than described as drag-and-drop. VISUAL SCRIPTING IS THE MECHANISM: USE VISUAL SCRIPTING TO DESIGN AI AGENT WORKFLOWS, AUTOMATING AND SEQUENCING ACTIONS, CONNECTING DATA SOURCES WITH FINE-TUNED MODELS, CHAINS AND PROCESSING FUNCTIONS. Sequencing actions and composing chains against specific fine-tuned models is genuine multi-step construction, and binding a step to a particular model is a level of control most visual builders in this lane do not offer. SKILLS ARE THE VENDOR'S NAME FOR PROCESS ORCHESTRATION TASKS, built in the low-code interface and combined with agents; independent analyst coverage describes the agents-plus-skills combination as providing robotic process automation capabilities. The vendor's own framing of ONTOLOGY-DRIVEN WORKFLOWS ties the orchestration layer to the knowledge graph, so a process step can be defined against business entities rather than against raw data. THE MULTI-AGENT LAYER IS DOCUMENTED as governable agentic orchestration coordinating multiple agents across departments, and the secondary Multi-agent platform categorisation on this record is consistent with that. THE ARCHITECTURAL POINT WORTH CARRYING is the split the vendor makes between assistants and robots inside one orchestration model: a workflow can hand a step to an autonomous agent or route it through a person, which is why the oversight cell and this one describe the same visual surface from different angles. What is not documented is control-flow vocabulary. No branching, looping, conditional or parallel construct is named on any page reached, and no failure or retry behaviour is described. Sequencing and chaining are documented; expressiveness beyond that rests on the visual scripting claim.

Stands at F but at medium confidence, and the reason is recorded. Multi-step orchestration is documented beyond doubt: a node graph with typed ports, loops, branching, parsing, type conversion, component grouping for reuse, and freeze to pin upstream state. Multi-agent is the half resting on thinner first-party material. What IS documented is that flows and components can be used as agent tools, so one agent invokes another flow, which is genuine composition. The stateful multi-agent LangGraph integration that would settle this decisively was described only in third-party coverage of the 1.8.4 release and was not confirmed on a Langflow page this pass, so it is not used here. Fetching the Agents section would take this to high confidence either way. Note also that Langflow now ships an ASSISTANT that generates flows and custom components from a prompt, which is a builder aid rather than runtime orchestration and is deliberately not credited on this axis.

Stands at P. The July basis called channel coverage limited and that reading holds. THE EVENT CLASS IS DOCUMENTED THROUGH THE WORK ITSELF. Agents are triggered by document and data ingestion, with the platform ingesting from any source or format including legacy databases, emails, scans and forms. For a document automation platform that is the primary and correct entry route: work arrives because a document did. A SECOND ROUTE IS THE API. An API-first architecture with agents embedded into the customer's applications means an external system can invoke an agent, which is the inbound programmatic class and is credited on Ext as a surface but counts here as a route by which work reaches an agent. WHAT IS ABSENT IS THE REST OF THE SPREAD. No scheduled, cron or recurrence capability is named. No inbound webhook or external event subscription appears as a distinct capability. And no channel layer exists at all: no chat, email, messaging or embedded conversational surface through which a person reaches an agent is documented, beyond Altilia Insights as an in-platform assistant. THAT IS COHERENT FOR THE PRODUCT rather than a deficiency, and the grade should be read that way. This is back-office document and knowledge automation for regulated enterprises, not a conversational front end. Work arrives as documents and results go into systems; there is no customer sitting on WhatsApp waiting for a reply. A vendor with five channels and no knowledge graph would score better on this axis and be worse at the job Altilia does. Confidence medium: no triggers, scheduling or automation page was reached across two passes, and a platform doing overnight batch document processing for thirty enterprise clients almost certainly schedules, so this is likelier undocumented than absent.

Stands at P, and the April grade turns out correct. Two of the three classes are well documented. Events: a dedicated Webhook component with its own POST endpoint per flow, API-key authenticated by default, described by the vendor as the versatile event-driven entrypoint and paired with a Parser for payload extraction. Channels: the Langflow API, an embeddable chat web component for any site, the Playground, and flows exposed as MCP tools that external assistants invoke. THE MISSING CLASS IS SCHEDULES. No scheduler, cron or recurring-run capability appears anywhere in the documented navigation, and the vendor's own trigger documentation covers only the API and webhooks. That is consistent with the product's design, since Langflow expects an external scheduler to call the webhook, but it is a genuine gap against this axis. Would move to F on a documented scheduling surface.

Stands at F and is the strongest cell on the record by a clear margin. It is also where several facts removed from other cells correctly land. THE KNOWLEDGE GRAPH IS THE PLATFORM'S CORE, not a retrieval add-on. The vendor states it AGGREGATES AND ORGANIZES ALL INFORMATION WITHIN A COMPANY, UNIFYING DATA FROM EVERY POSSIBLE SOURCE, STRUCTURED, UNSTRUCTURED, AND EVEN PHYSICAL DOCUMENTS, INTO A SINGLE COHERENT DATABASE. A graph is the strongest form of maintained retrieval structure there is, because it encodes relationships between entities rather than similarity between text chunks, and it is queryable by traversal rather than only by nearest neighbour. THE ARCHITECTURE IS HYBRID AND THAT IS THE DIFFERENTIATOR. Large and small language models are combined with the graph in a neuro-symbolic design, described as GRAPHRAG WITH TRAVERSAL REASONING. The symbolic half is what makes an answer traceable to a structure rather than to a probability. INGESTION BREADTH IS EXCEPTIONAL AND NOW SITS HERE rather than on Int: any source or format, including long complex documents, standardised forms, SCANNED IMAGES OR HANDWRITTEN NOTES, emails and legacy databases. Handwriting and scans matter for the regulated European buyers this targets, where the source of truth is frequently paper. CITATION IS DOCUMENTED, which few records in this lane manage: users interrogate the knowledge base in natural language and receive PRECISE, FACT-BASED ANSWERS WITH SOURCE ATTRIBUTION. Attribution is what makes a grounded answer checkable, and it is the property that carried over from Obs where it did not belong. RAG techniques retrieve and fine-tune LLMs and SLMs against enterprise knowledge, and the vendor reports above ninety-five percent accuracy on domain tasks, which is vendor-reported and recorded as such.

Stands at F. Retrieval is the capability Langflow was built around and it is assembled from first-class parts rather than behind one switch: document loaders, text splitters with configurable chunk size and overlap, embedding models, vector stores as provider bundles, and retrieval components, all connected on the canvas with typed ports. Relevant to the open Knowledge convention, this vendor sits on the MAINTAINED STRUCTURE side but exposes the construction rather than the result, which is a third shape worth naming: the customer builds and owns the index, chooses the store, and can point at a database they already run. Because vector stores are bundles rather than a proprietary layer, the knowledge base can live entirely in customer infrastructure, which pairs with the Dep grade. File management for uploaded documents is documented separately with metadata tracked in the database.

Stands at P under the state limb of the 31 August ruling, and the interesting part is a distinction the July basis blurred. CONTINUOUS LEARNING IS NOT MEMORY, and this record is the clearest case for separating them. The vendor documents that THROUGH CONTINUOUS LEARNING AND HUMAN FEEDBACK, AGENTS BECOME INCREASINGLY ACCURATE, and that reviewed corrections feed improvement over time. Retaining corrections across runs is the accumulation limb of the ruling, and on the wording alone this would reach Full. BUT THE MECHANISM IS FINE-TUNING, NOT STATE. The platform page describes creating TRAINING DATASETS from real documents and using RAG techniques to RETRIEVE AND FINE-TUNE LLMS AND SLMS. Corrections are absorbed by updating model weights offline, not written to a store the agent reads at runtime. Those are genuinely different capabilities: a fine-tuned model is better at the task in general; a memory lets an agent recall what happened with this customer last month. Crediting weight updates as memory would let every platform with a feedback loop claim the axis. WHAT IS LEFT IS THIN. Altilia Insights is a conversational assistant, so session context exists within a conversation, which is the ruled definition of Partial. The Knowledge Graph persists and is addressable, but it is the grounding structure and is credited on Know; treating a corpus as agent memory would be the same double-count. NO CROSS-SESSION AGENT STATE IS DOCUMENTED: no memory store, no retention policy, no per-user context, no primitive by which an agent writes state in one run and reads it in another appears on any page reached across two passes. WORTH CARRYING FOR THE LANE: this is the second record this session where a learning loop presented as memory. The test that separates them is whether the artefact is read at inference time or absorbed into weights.

N>F, and the N was badly wrong rather than marginally wrong. Memory is not merely present, it is ON BY DEFAULT: the vendor states that in any project using Chat, memories are always being stored. The Agent component carries built-in chat memory enabled by default and retrieves messages from previous conversations on the same session ID, which is cross-session persistence in the plain sense the axis means. Three properties push this to F rather than P. Sessions are first-class and arbitrary, so using user IDs as session IDs isolates each user's history. Storage is pluggable to external databases rather than trapped in the product. And memory is INSPECTABLE AND EDITABLE from the Playground, where a builder can read, edit and delete stored messages to change how the agent behaves, which is a rarer guarantee than persistence itself. Note for the axis generally: Langflow REMOVED the Memory component category in version 1.5 and folded it into Helpers and Bundles, so a grader searching for a memory category would find nothing. That is very likely how this cell came to be N, and it is a trap worth remembering.

Stands at F, and the mechanism is documented as a workflow stage rather than as a governance principle. THE REVIEW STEP IS EXPLICIT: REVIEW AND VALIDATE DATA CLASSIFICATIONS, EXTRACTIONS AND AI-GENERATED ANSWERS, with the vendor adding INTEGRATE THE FEEDBACK OF HUMAN EXPERTS TO ENHANCE ACCURACY AND CONTROL AT KEY STAGES. At key stages is the phrase that matters: the checkpoint sits at chosen points in the workflow rather than being a blanket setting, which is the same property that earned agentx its Full earlier today. THE ARCHITECTURAL DISTINCTION IS THE STRONGER EVIDENCE. Altilia ships agents in two declared forms, ASSISTANTS that support a human decision and ROBOTS that automate a process autonomously. A platform that makes the autonomy level a first-class choice at design time has built oversight into its model of what an agent is, rather than bolting an approval toggle onto an autonomous default. FOR THIS PRODUCT THE VALIDATION STATION IS THE RIGHT SHAPE. The work is document classification and extraction feeding downstream processes, so the consequential moment is accepting an extracted value, not calling an external tool. A human confirming a classification before it enters the knowledge graph is the approval gate that matters here, and it is where the vendor put it. THE FEEDBACK LOOP IS DOCUMENTED AS CLOSING: reviewed corrections feed continuous learning, so oversight improves the system rather than only catching individual errors. What is not documented is a policy layer: no confidence threshold routing low-certainty items to review, no action allowlist and no spend or scope limit appears on any page reached. Oversight here is placed by design rather than triggered by rule.

N>F, and unlike Mem this one looks like genuine PRODUCT CHANGE rather than a grading error: HITL is documented as a Langflow 1.11 capability and the record was built in May against roughly 1.8, so N may well have been correct at build time. That distinction matters for the lane note, because it separates records that were graded badly from records that simply aged. Two independent mechanisms earn the F, and the second is unusually granular. The Human Input component pauses at a chosen point and opens one branch per configured user action, so approve and reject route to different downstream paths rather than merely gating. Agent tool approval is finer still: Requires approval is set PER TOOL, so a run pauses only when the agent reaches the specific dangerous tool, which is a more precise control than the whole-run gates most vendors in this lane ship. The checkpoint-and-resume design is also worth noting, since completed steps are not re-executed on resume.

F>P, and the vendor's own security paragraph is what decides it. The July basis read the certifications as held; the sentence says something weaker. THE EXACT WORDING: Altilia ensures robust data security through its Integrated Management System, ALIGNED WITH ISO/IEC 27001, ISO/IEC 27017, AND ISO/IEC 27018 STANDARDS. Aligned with is the second rung of the hedge ladder in section 7 and is named there explicitly as Partial. It says the management system was built to those standards; it does not say a certification body audited and issued against them. THE SELF-CONTRADICTION IS INSIDE ONE PARAGRAPH AND ACROSS TWO PAGES. The same sentence continues THESE CERTIFICATIONS VALIDATE, treating alignment as certification, and a separate page describes an ISO-CERTIFIED PLATFORM. Under section 7 the more precise and technical statement wins over the looser marketing one, and the precise statement is the one that names the management system and the three standard numbers. WHY THIS MATTERS RATHER THAN BEING PEDANTRY: for a vendor selling data sovereignty to European regulated buyers, whether an ISO certificate exists is the first procurement question, and the distinction between aligned and certified is exactly what a procurement team is checking. Recording it as certified when the vendor says aligned would put a claim on the index the vendor itself does not make. THE CONTROL HALF IS GENUINELY MET, which is why this is Partial and not lower: ACCESS CONTROLS, ROLE-BASED PERMISSIONS is named, alongside risk management, incident response, and continuous monitoring and audits. Under the conjunction bar, one half documented and one half asserted is Partial. Two facts moved off this cell as belonging elsewhere: data sovereignty and deployment location are graded on Dep, and customer IP ownership of models is a commercial term rather than a security control. No certificate number, issuing body, scope statement or trust page was reached. That is the single check that would move this to Full.

Stands at P, and P is the right shape for a self-hosted open-source product rather than a criticism of it. Real controls exist and are documented: API-key authentication on every flow endpoint, webhook authentication enabled by default with an explicit warning against disabling it, automatic stripping of API keys and tokens from saved flow data, and encrypted storage of component credentials in a variables table. What cannot be present is the other half of the section 7 conjunction. There is no attestation, and there structurally cannot be one that covers a customer's own self-hosted instance, since the operator runs the infrastructure. That is a property of the delivery model, not a deficiency to hold against the vendor, and it is the reason this sits at P rather than N or F. ONE THING A BUYER SHOULD KNOW, recorded because it is first-party: the project's own README carries a standing caution to upgrade past a named remote-code-execution CVE, and third-party coverage reports further unauthenticated RCE-class CVEs during 2026. A visual builder that executes arbitrary Python is a large attack surface and should not be exposed to untrusted networks.

F>P. The July basis cited AI ops management, monitoring hundreds of agents, debugging and observability, and explainable results. Reading the platform page against those, two of the four are the wrong subject and one belongs to another axis. WHAT THE MONITORING ACTUALLY WATCHES IS MODELS, NOT AGENT RUNS. The documented capability is MONITOR AND TRACK MODELS PERFORMANCE TO SPOT INEFFICIENCIES and EASILY MONITOR AND MAINTAIN UP-TO-DATE AI MODELS THROUGHOUT THEIR LIFECYCLE, with automated resource adjustments based on performance and workload. That is model operations, an accuracy-and-lifecycle discipline, and it is a real capability. It answers whether the model is performing well; it does not answer what a given agent did on a given document and why. EXPLAINABILITY BELONGS TO KNOW AND HAS BEEN MOVED. Verifiable answers with SOURCE ATTRIBUTION come from the Knowledge Graph and are a property of the grounding layer. Citing a source is not a record of execution, and counting it here would let the same architectural fact carry two axes. WHAT IS MISSING IS THE WHOLE OF THE AUDITING HALF. No run history, per-document processing record, agent action log, decision trace, retention period or export path appears on any page reached across two passes. THE GAP IS SHARPER THAN THE GRADE SUGGESTS FOR THIS BUYER. Altilia sells to regulated European enterprises processing licences, tax forms, tenders and public registries, and positions on deterministic auditable AI. Those are exactly the workloads where someone will later ask why a document was classified as it was. Model-level accuracy metrics do not answer that for an individual case. PARTIAL RATHER THAN NONE because model performance tracking is real, is customer-facing, and does tell an operator that something has degraded, which is more than several records in this batch offered.

Stands at F but confidence drops to medium, because what I verified is run inspection rather than a trace product. Per-component Inspect showing output and logs is genuinely the why rather than the what, since a builder can see what each node in the graph produced and where a run diverged, and that is more than the reporting dashboards several vendors are graded F for. Monitor endpoints expose stored messages programmatically. WHAT I DID NOT VERIFY FIRST-PARTY this pass is the external tracing integration with LangSmith and Langfuse that the record's own prose claims and that third-party coverage repeats; it is not used to support this grade. Also not verified: any audit trail of who changed a flow, which for a self-hosted open-source tool would ordinarily be the operator's responsibility rather than the product's. Fetching the Develop section would settle both. If neither is there, this cell is arguably P.

Stands at F and is among the better-evidenced Dep cells in the lane, because deployment control is the company's positioning rather than an enterprise-tier add-on. THREE MODES ARE DOCUMENTED CONSISTENTLY across the home, why-altilia and platform pages: SaaS, private cloud, and on-premises, described by the vendor as DEPLOYMENT SOVEREIGNTY. Repetition across pages matters here; several records this session rested a deployment claim on one sentence in one place. THE PART THAT MAKES IT SUBSTANTIVE RATHER THAN A HOSTING MENU is that the models travel with the deployment. Domain-tuned SLMs and LLMs are SERVED LOCALLY, so an on-premises customer is not running a thin client that ships documents to a hosted model for inference. Model inference is the leak in most agent platforms' residency story, and this one closes it by design. For a buyer processing tax forms and civil registries, that is the difference between a usable option and a nominal one. DATA SOVEREIGNTY IS STATED AS THE PURPOSE, with data kept within the customer's secure infrastructure and customers retaining ownership of data, models and assets. The company's whole market position is a sovereign European operating system for agentic AI competing against US hyperscalers, which is only coherent if the deployment claim is real. Per the 30 August ruling the sovereignty and residency properties are graded here alone and are deliberately not credited again on Sec, where the attestation now sits at Partial on its own evidence. What is not documented is region selection within the SaaS offering, installation requirements, or which components run customer-side in the on-premises mode.

Stands at F. For an MIT-licensed self-hostable product the residency question answers itself in the strongest possible way: the customer runs the software, so the region, the database and the perimeter are all theirs by construction, and there is no vendor-side data location to negotiate. That is a categorically different guarantee from a SaaS vendor offering region selection, and it should be described that way on comparison pages rather than levelled flat with them. The delivery surface is unusually wide: a documented deployment section, containerised deployment, a Python package, a signed desktop application for macOS and Windows for people who do not want to manage environments at all, and LFX for embedding Langflow as a dependency inside another application. A no-op database mode even allows running with no persistence. Recorded rather than graded: IBM Langflow Cloud is reported by third parties to be in private preview, so the managed path is not yet generally available and the production path today is infrastructure the customer stands up.

Stands at F under the 31 August bar, which asks whether the customer receives packaged assets ready to adopt. THE PREBUILT SET IS NAMED BY BUSINESS DOCUMENT RATHER THAN BY CAPABILITY, which is the stronger form. The vendor documents DEPLOY READY-TO-USE AGENTS FOR COMMON USE CASES SUCH AS EXTRACTING DATA FROM INVOICES, ORDERS, AND TRANSPORT DOCUMENTS. An invoice agent is a finished job; a toolkit is a starting point, and the difference is exactly what this axis measures. A SECOND, BROADER SET IS DOCUMENTED BY FUNCTION: specialised agents for document classification by semantic context, data extraction from unstructured sources, semantic search and question answering, summarisation into reports, and document generation for emails and operational content. Those are role-shaped rather than component-shaped. THE TWO-ROUTE STRUCTURE IS EXPLICIT AND IS WHY THIS CLEARS FULL RATHER THAN SITTING WHERE THE GENERATION-ONLY RECORDS LANDED THIS SESSION. The vendor states customers either deploy ready-to-use agents or use the IDE to create bespoke ones, so adoption of a packaged asset is a first-class path beside building, not a fallback. Kalcend dropped to None earlier today because generation from a prompt had replaced the catalogue entirely; here both exist. BLUEPRINTS FOR COMMON USE CASES sit alongside, and the vendor's positioning of a digital workforce implies role-shaped packaging rather than parts. What I did not reach is an enumerated gallery, so the size of the ready-to-use set rests on the named examples plus the vendor's description. Three named document types is thin as a catalogue but specific enough to be checkable, which is more than an unenumerated count would be.

Stands at F, and under the 31 August ruling this is a clean case rather than a marginal one. The vendor states that Langflow includes several pre-built templates ready to use or customise, and named templates recur throughout the documentation as the assumed starting point, with instructions repeatedly opening a flow FROM a template rather than from blank. That is packaged assets the customer adopts, which is the operative bar, and the browsable-gallery question does not arise. Two further layers sit alongside: grouped components can be saved to the component menu as reusable custom components, and flows import and export as JSON so a team or the community can circulate finished assets. Not counted here, deliberately: the Langflow Assistant generates flows and components from a prompt, which is generation rather than a supplied asset, and crediting it would blur the axis.

Stands at F, and it is unusually well documented for this axis because model choice is the vendor's strategic position rather than a feature. THE BREADTH IS THE FULL RANGE: any commercial or open-source model, large and small, with the vendor's own framing being YOUR MODELS, YOUR CONTROL. Customers retain IP ownership of models trained on their data. THE SMALL-MODEL AND LOCAL-EXECUTION HALF IS THE DISTINCTIVE PART and separates this from a provider menu. Domain-tuned SLMs and LLMs are SERVED LOCALLY, with model selection driven by COST AND DATA TYPE. Choosing a small local model for a high-volume document class and a large hosted one for hard reasoning is a genuine economic lever, and running the small one inside the customer's own infrastructure is what makes it available to buyers who cannot send documents to a hyperscaler at all. FINE-TUNING COMPLETES IT: the platform retrieves and FINE-TUNES LLMS AND SLMS against enterprise knowledge, so the customer is not only choosing among models but shaping the one they choose. Few records in this lane document that. THE STRATEGIC CONTEXT IS WORTH CARRYING because it explains why this cell is strong rather than incidental. Altilia positions as a sovereign European alternative to US hyperscalers, and model portability is the technical substance behind that claim: a platform that could only call OpenAI could not make it. Per the standing convention, local model execution is credited here as customer model control and the deployment location it implies is graded on Dep, not counted twice. What is not documented is a named provider list, a model picker interface, or automatic routing between models at runtime, none of which this axis requires.

Stands at F. Model choice is a per-node property here rather than an account setting, so different steps of one flow can run on different providers, and the Language Model component's provider and model fields are the ordinary way to configure any LLM step. Local models via Ollama matter for the same reason Dep does: combined with self-hosting they let the entire stack run inside the customer's boundary with no external inference at all, which very few vendors in this lane can offer. Provider bundles mean the roster grows by integration rather than by vendor permission, and custom Python components mean an unsupported provider is a component away. Not to be confused with the Ext credit for exposing flows as MCP servers, which is the opposite direction of travel and is graded there.

Stands at F under Mike's 30 August bar, though it is the least verified of the Full cells on this record and the note should say so. WHAT SUPPORTS IT. The vendor states an API-FIRST ARCHITECTURE, and the direction is the Ext one: agents are CONNECTED DIRECTLY INTO YOUR OPERATIONS AND APPLICATIONS and EMBEDDED INTO YOUR ENTIRE TECH STACK, meaning the customer's own systems invoke Altilia agents rather than the reverse. API-first is a claim about how the platform is built rather than a feature bolted on, and every capability being reachable through the API is what the term means. SIMPLE SDKS ARE NAMED AS A FIRST-CLASS BUILD ROUTE, listed alongside natural language and visual scripting as the three ways a customer creates agents. A vendor offering an SDK as one of three peer entry points is describing a developer surface, not an afterthought. CONFIDENCE IS MEDIUM AND THE GAP IS THE SAME ONE I NAMED ON AGENTX TODAY: no API reference, endpoint list, authentication documentation or SDK package was reached across two passes. The API is documented as an architectural property and a deployment route rather than through a developer surface a reader can inspect. I have graded consistently with agentx, where one-click-to-API carried Full on the same basis, rather than applying a stricter test here. No MCP server was found, which under the ruling does not withhold the grade. THE ASYMMETRY WORTH RECORDING: this vendor documents its knowledge architecture in unusual depth and its developer surface barely at all, which is consistent with selling to business buyers through enterprise engagements rather than to developers. A developer portal is the natural check at lane close, alongside the connector catalogue for Int.

Stands at F and clears Mike's 30 August Ext bar on multiple routes. The distinguishing property is that extensibility runs BOTH WAYS through MCP, which few vendors in this lane manage: Langflow exposes flows as an MCP server so external clients call them as tools, and connects outward to external MCP servers as a client. Under section 7 the server direction credits Ext and the client direction credits Int, and they are graded separately here rather than one fact doing double duty. Beyond MCP, every flow is callable over a documented REST API, flows import and export as JSON so an agent definition is a portable artifact, custom components are authored in Python, and an embeddable chat web component ships for front-end integration. LFX is a separately documented package for running flows as a dependency.

Stands at P under the 31 August Eval bar. The July basis was accurate and the refinement here is about which half is present. THE MEASUREMENT SURFACE IS REAL BUT AIMED AT MODELS. The vendor documents MONITOR AND TRACK MODELS PERFORMANCE TO SPOT INEFFICIENCIES, ENABLE CONTINUOUS LEARNING AND IMPROVEMENT OVER TIME, alongside human review and validation of classifications, extractions and generated answers. Reviewed corrections both catch individual errors and feed the training loop, so quality is measured and acted on. WHY IT DOES NOT REACH FULL. The ruled bar asks for a result the customer can read AND COMPARE about the agent's behaviour on their own work. Nothing documents a test set held aside, expected outputs recorded, a scored run, or a comparison between one version of an agent and another. Model performance tracking tells an operator that accuracy has moved; it does not let them establish that a specific change caused it, which is the question a regression harness answers. THE VENDOR-REPORTED ACCURACY FIGURE IS RECORDED AND NOT CREDITED. Above ninety-five percent on domain-specific tasks is a vendor benchmark, and under the standing convention vendor self-benchmarking is Partial-class evidence at best, the same handling applied to codebuff. THE STRAIN WORTH NAMING for a platform of this kind: the review station generates exactly the material a golden dataset needs, since every human validation is an expected output recorded against a real document. Turning that into a scored regression suite is a small step from what exists, and its absence is more likely undocumented than unbuilt. Confidence medium; no evaluation, testing or debugging page was reached across two passes, and the debugging dashboard referenced in the July note was not found on any page this pass.

Stands at P. The debugging surface is genuinely good and better than several vendors graded higher elsewhere: a Playground for interactive testing, single-component runs that execute one node in isolation without its upstream dependencies, per-component output and log inspection, freeze to pin upstream state while iterating on one node, editable and deletable message logs so a tester can reshape history and observe the effect, and backup flows created automatically before breaking component updates. What is absent is the F bar as this lane has set it. There is no evaluation product: no scoring, no test-case generation, no regression suite, no benchmarking, nothing that measures one version against another. Compare lyzr, which earns F on an LLM-as-a-judge evaluator with production-readiness scoring, and stackai on the same shape. This is debugging, not evaluation, and the distinction is the axis. Recorded honestly: version 1.12 is in preview and this is a fast-moving open-source project, so the cell is worth re-checking each sweep.

Stands at N, and confidence rises to high because a genuine near-miss was found and refused rather than nothing being found. THE NEAR-MISS IS THE STRONGEST ON THIS RECORD AND WOULD CATCH A GRADER SKIMMING. Altilia calls its autonomous agents ROBOTS, distinguishing them from assistants, and independent analyst coverage states that the combination of agents and skills PROVIDES ROBOTIC PROCESS AUTOMATION CAPABILITIES. A vendor whose own vocabulary is robots and RPA, in a market where RPA historically means screen automation, reads at first glance like the positive case for this axis. IT IS NOT, AND THE DISTINCTION IS THE ONE THAT MATTERS FOR THIS LANE. Altilia's robots are autonomous agents that classify, extract, summarise and generate against documents and a knowledge graph, reaching systems through an API-first architecture and connectors. RPA capabilities here means the outcome replaces the manual work RPA was bought for, not that the mechanism is screen driving. No browser control, navigation, form filling, screen interaction, visual grounding or desktop automation appears on any page reached across two passes. COMPUTER VISION IS THE SECOND NEAR-MISS AND IS ALSO REFUSED. The platform uses computer vision and document analysis and recognition to read scanned images and handwritten notes. Reading a document image is intake, credited on Know, and is the opposite of operating an interface: the page is being interpreted, not driven. THE ABSENCE IS STRUCTURAL. A platform whose entire premise is turning unstructured sources into a queryable knowledge graph reaches data by ingestion and API, which makes screen automation unnecessary. WORTH CARRYING TO ENTERPRISE OPERATIONS, where IDP and RPA-adjacent vendors will recur: robots in the vendor's vocabulary and RPA in an analyst's are not evidence of computer use, and the test is whether an interface is being driven.

P>N, the one downward correction and the same axis error the Coding agent lane corrected thirteen times, appearing here in a different disguise. Nothing first-party drives a browser. What exists is a Web Search component that the vendor's own documentation describes as WEB SCRAPING subject to rate limits, plus an API Request component, plus provider bundles. Fetching pages over HTTP is not operating software that lacks a programmatic interface. Two near-misses were checked and refused. The Apify bundle brokers a hosted RAG Web Browser actor, but under section 7 that is Apify's capability reached through an integration, not Langflow's, in the same way Zapier exposing Motion actions was not Motion's extensibility. And community projects wiring the browser-use library into Langflow as a custom component are third-party work, not shipped product; the fact that this requires a custom component is itself evidence of absence. Confidence is medium because the component catalogue is large and version 1.12 is already in preview, so a first-party browser component could appear.