Akro AI vs ReN3 (2026)

Stands at P, re-based, with the source classes now named first party rather than summarised from a funding release. WHAT IS DOCUMENTED IS BREADTH: the platform CONNECTS DOCUMENTS, EMAILS, WEBSITES, DATABASES, ENTERPRISE APPLICATIONS, AND MULTIPLE FILE FORMATS. Databases and enterprise applications alongside content sources are genuine integration classes, and access rights are synced from those source systems so the connection carries permissions rather than bypassing them. WHY IT HOLDS AT PARTIAL. Not one integration is named anywhere on the site. No connector catalogue, no count, no marketplace and no integrations page exists in a navigation that was read in full. Enterprise applications is a category, not evidence of breadth across classes, and the same reasoning held ai-library and altilia at Partial earlier in this lane. ONE LIST THAT LOOKS LIKE EVIDENCE AND IS NOT, worth naming because it is prominent and easy to miscount. The homepage carries eleven logos under IN PARTNERSHIP WITH: Red Hat, Dell, NTT, AWS, Microsoft, Lenovo, Nutanix, Google, Alibaba, Cisco and Capgemini. Those are infrastructure, hardware and systems-integrator channel partners, the route by which this platform is sold and deployed. They are not applications an agent connects to and calls as tools. Reading a channel roster as an integration catalogue would inflate this cell substantially. THE TOOL-CALLING HALF IS UNDOCUMENTED. Nothing describes an agent invoking an external system as a tool during execution; the documented action set is retrieval, analysis, generation and report production over the knowledge layer. THAT SHAPE IS COHERENT FOR WHAT THIS IS. A sovereign platform whose premise is that data never leaves the customer's environment has a structural reason to pull sources into a governed knowledge layer rather than to reach outward into third-party SaaS at run time. Partial describes it accurately.

Stands at P, re-based onto the product page, with the roadmap half of the July basis now confirmed rather than assumed. WHAT IS DOCUMENTED. Business users CREATE AI AGENTS USING INTUITIVE NO-CODE TOOLS, ENABLING TEAMS TO AUTOMATE REPETITIVE WORK, GENERATE REPORTS, REVIEW DOCUMENTS, SUMMARIZE INFORMATION, AND EXECUTE ENTERPRISE WORKFLOWS WITHOUT WRITING CODE. Executing enterprise workflows is multi-step work and the platform is described as combining AI agents, knowledge management, governance and orchestration in one platform. ONE TERM NEEDS SEPARATING CAREFULLY, because it would otherwise read as evidence for this cell. ADAPTIVE AI ORCHESTRATION on this site means orchestrating MODELS, not agents: the product page's orchestration passage is the one about routing each request across multiple AI models by complexity, latency, policy and cost. That is credited on Model. Reading a model-routing claim as agent orchestration would be the one-fact-wrong-axis error, and the vendor's own wording makes the distinction easy to miss because both use the word orchestration. WHY IT HOLDS AT PARTIAL. The agentic workflow builder that would chain agents into multi-step auditable processes is roadmap, confirmed by the vendor's own May 2026 funding announcement naming it as a use of seed proceeds alongside the agent marketplace. Neither has shipped as of this pass. No branching, conditional, looping or parallel construct is named, no multi-agent coordination or delegation is described, and no visual flow surface appears on any page reached. SO THE UNIT OF EXECUTION IS THE AGENT, not a composed process. A customer builds agents that do multi-step work individually; composing them into a governed pipeline is the thing being funded rather than the thing being sold. This is the cell most likely to move on the next cycle, and the timing is knowable: seed money was raised in May 2026 specifically to build the workflow builder.

Stands at F, re-based from the funding announcement onto the product page, and the source breadth is wider than the format count conveyed. THE UNIFIED KNOWLEDGE LAYER IS THE PLATFORM'S FIRST CLAIM, not a supporting feature: ReN3 CONNECTS DOCUMENTS, EMAILS, WEBSITES, DATABASES, ENTERPRISE APPLICATIONS, AND MULTIPLE FILE FORMATS, CREATING A UNIFIED KNOWLEDGE LAYER THAT AI CAN SECURELY UNDERSTAND AND RETRIEVE FROM. Emails and databases alongside documents is the combination that matters for the back-office work this sells into, where the answer to a procurement question lives in a contract, a thread and a ledger at once. THE FORMAT EXAMPLES ARE MORE TELLING THAN THE COUNT. The product page names PDF, SPREADSHEET, CAD DRAWING, OR SCANNED IMAGE. CAD is the one to carry: it is not a format anyone claims casually, it is specific to the construction and manufacturing verticals the site sells to, and it sits alongside scanned images and the audio and video support named in the funding release. This is genuine multimodal ingestion rather than a document indexer with a wide file filter. CITATION IS DOCUMENTED, which few records in this lane manage. The homepage states COMPREHENSIVE AUDIT TRAILS AND CITATION-BACKED ACCURACY GROUNDED IN YOUR ENTERPRISE KNOWLEDGE, and the product page adds that EVERY AI RESPONSE IS GROUNDED IN YOUR ENTERPRISE KNOWLEDGE and every answer is GROUNDED IN TRUSTED ENTERPRISE DATA. Citation is what makes a grounded answer checkable by the person receiving it rather than merely asserted. THE PERSISTENCE LINE IS CLEARED by the knowledge layer being a maintained structure the platform builds and retrieves from, built on an enterprise content management core with access rights synced from source systems, so retrieval respects the permissions of the asking user. One limit unchanged from July: no chunking, embedding model, index refresh or retrieval configuration is documented anywhere on the site.

P>F. The July basis said EXPLICIT HUMAN APPROVAL WORKFLOWS NOT DETAILED and rested Partial on cross-model verification and audit trails, which are accuracy controls rather than oversight. Approval workflows are named first party on the product page. THE GOVERNANCE SENTENCE IS THE EVIDENCE, and it is specific about scope before it is specific about mechanism: EVERY WORKSPACE, AI AGENT, USER, AND CONNECTED DATA SOURCE IS GOVERNED THROUGH CENTRALIZED POLICIES, ROLE-BASED PERMISSIONS, APPROVAL WORKFLOWS, AND COMPREHENSIVE AUDIT TRAILS. Approval workflows sit as one of four named mechanisms, applied across agents rather than offered as a per-workflow option. THE SCOPE CLAUSE IS WHAT MOVES THIS PAST THE RECORDS HELD AT PARTIAL THIS LANE. wassist and aigensei documented escalation, which is an exit; snaplogic documented guardrails with no human in them. Here approvals are named as governing every AI agent, which is the checkpoint class rather than the exit class, and it is stated as a property of the platform rather than something a builder assembles. THE SUPPORTING LAYER IS REAL AND UNUSUALLY WELL MATCHED TO THE BUYER. Multi-tier access controls are built into every AI interaction, agent execution is described as explainable, and the platform is sold into government, banking, insurance and healthcare where an unreviewed agent action is the thing procurement asks about first. CONFIDENCE IS MEDIUM AND THE LIMIT IS PRECISE: approval workflows are named, not described. Nothing states where an approval sits in an agent run, who is assigned, what happens on rejection, or whether the gate is configurable per agent. That is the difference between this and agentx, which documents checkpoints placeable at any step and holds Full at high confidence. Cross-model verification stays recorded as an accuracy technique and is not counted here; it constrains output quality, not agent authority.

F>P under the 1 September ruling, and the correction has two parts: the grade, and a published claim that had to come out of the prose regardless of the grade. WHAT THE VENDOR ACTUALLY SAYS. The badge in the Accredited By block on the homepage and in the footer reads SOC2-COMPLIANT. The asset is named f-SOC2-Compliant.png. There is no type, no scope, no auditor, no report and no trust centre anywhere on the site. Alongside it sit an IMDA accreditation and a Meta Business Partner badge. THE INDEX WAS PUBLISHING A CLAIM THE VENDOR HAS NEVER MADE. The description read SOC 2 TYPE 2 CERTIFIED, which came from technotrenz, a trade blog. Of the two excluded sources behind this cell, ai-market-watch said SOC2-compliant and was right; technotrenz said Type 2 certified and was wrong. The description is corrected. Removing an unsourced certification claim is not asserting the opposite and carries none of the false-negative risk that made me reluctant to move the grade. WHY PARTIAL RATHER THAN FULL. A compliance badge with no type, scope or auditor is a hedged attestation, and hedged forms sit at Partial with the hedge written into the note. That is the same ladder that placed altilia at Partial on ALIGNED WITH ISO 27001 and aigensei at Partial on SOC 2-ALIGNED, and one rung below joget's ISO/IEC 27001:2022 CERTIFIED and hostinger's viewable certificate. WHY PARTIAL RATHER THAN UNKNOWN. First party evidence is reachable and clears the P floor on its own: the homepage states GOVERNANCE AND MULTI-TIER ACCESS CONTROLS BUILT INTO EVERY AI INTERACTION and COMPREHENSIVE AUDIT TRAILS AND CITATION-BACKED ACCURACY. Unknown is for a cell whose only support is an excluded source with no first party evidence reachable. That is not this. WHAT WOULD MOVE IT TO FULL: a stated type and scope, or a named auditor, or a report path, alongside the access controls already documented. government.php and bank-financial.php are the likeliest places for a regulated-sector buyer to be given that detail.

P>F. The July basis had visible audit trails and knowledge versioning with DEDICATED OBSERVABILITY TOOLING NOT DETAILED. The product page names observability directly and, more importantly, names the why rather than only the what. THE PHRASE THAT DECIDES IT IS EXPLAINABLE AGENT EXECUTION. The product page reads FROM EXPLAINABLE AGENT EXECUTION TO REAL-TIME AI COST MONITORING, ORGANIZATIONS MAINTAIN COMPLETE OPERATIONAL VISIBILITY. This axis turns on reporting versus auditing, seeing what happened versus reconstructing why, and explainable execution is a claim about the second. It sits directly after the governance sentence, so the object being explained is the agent's run rather than a model output. THE SECOND HALF IS CITATION-BACKED ACCURACY, stated on the homepage alongside comprehensive audit trails. Citations make an answer checkable against the source it came from, which is the document-workflow equivalent of a tool-call record: a reader can verify the chain rather than trust the conclusion. OBSERVABILITY IS NAMED AS A PLATFORM PROPERTY in the Optimized for Scale block, CONTROL AI COSTS WITH ADAPTIVE ROUTING, BUDGET GOVERNANCE AND OBSERVABILITY, and real-time AI cost monitoring is called out separately. Per-interaction cost visibility is unusual in this lane and matters here because routing is automatic: a customer who does not choose the model still needs to see what the choice cost. THE GOVERNANCE SCOPE CARRIES OVER: audit trails cover every workspace, AI agent, user and connected data source, so the record is agent-level rather than platform-level. CONFIDENCE IS MEDIUM. No retention period, export path, SIEM integration or example trace is documented, and explainable is asserted rather than illustrated. The Obs cells reviewed at Full in this lane on autogpt and retool both rest on a named artefact a customer can query; this rests on a named property.

Stands at F, now on a dedicated deployment page rather than on the funding announcement, and it is the strongest cell on the record. THREE MODELS ARE PUBLISHED AS THE PRICING STRUCTURE ITSELF, which is unusual: on this site deployment is not a feature within plans, it is what the plans are. BRING YOUR OWN CLOUD deployed in the customer's own cloud, MANAGED PRIVATE CLOUD as private tenancy on ReN3 cloud, and ON-PREMISE deployed in onsite or air-gapped servers and described as FULLY 100% AIRGAPPED. The product page adds public cloud and sovereign to the same list. THE MODEL HOSTING LINE IS THE DETAIL THAT MATTERS and it closes the loop most vendors leave open. Under BYOC, LLM HOSTING CAN BE GPU/API-BASED; under On-Premise, LLM HOSTING IS GPU BASED. So in the air-gapped configuration inference runs on customer GPUs by design, not as an option. Model inference is the leak in nearly every residency claim in this lane, and here the deployment page states where it runs as part of the plan definition. FULL FEATURES ON EVERY MODEL is stated explicitly on all three. That matters against the pattern seen repeatedly this lane, where self-hosting exists but arrives stripped of the governance features that made the platform worth buying. Here the tiering is on deployment shape and support, not on capability. EVERY OPTION IS A CUSTOMER-CONTROLLED OR PRIVATE-TENANCY ONE. There is no multi-tenant shared offering at all, which is coherent for a vendor selling to government, banking, insurance and healthcare in Southeast Asia and consistent with the zero external dependencies claim. Commercials recorded rather than graded: annual subscription with concurrent-user per site licensing on all three, ReN3 credits for AI workloads bundled into the managed cloud plan, and 24x7 managed cloud monitoring on that tier only.

Stands at N, and confidence rises to high because the marketplace is confirmed as roadmap by the company itself rather than merely unfound. THE FUNDING ANNOUNCEMENT STATES THE USE OF PROCEEDS: ReN3 will INVEST IN FURTHER PRODUCT DEVELOPMENT INCLUDING ITS AGENTIC WORKFLOW BUILDER AND UPCOMING AGENT MARKETPLACE. Upcoming is the operative word, and the lead investor's separate piece repeats that capital will ACCELERATE DEVELOPMENT of the workflow builder and UPCOMING AGENT MARKETPLACE. Independent coverage describes it as a PLANNED agent marketplace. UNDER SECTION 7 ROADMAP IS NOT SHIPPED, and a capability named as a destination for seed funding is as clearly future-tense as this axis encounters. Grading it would credit a company for what its investors are paying it to build. WHAT THE CUSTOMER RECEIVES TODAY is the ability to build agents rather than a library of agents to adopt. The platform is described as a no-code builder for deep vertical agents across administration, legal, finance, procurement and HR, and those are the functions a customer builds for, not a set of prebuilt packs. The vendor's own positioning reinforces this: the ability to BUILD AGENTS IN-HOUSE, WITHOUT DEPENDING ON A CENTRAL DATA SCIENCE TEAM OR A SINGLE VENDOR'S ROADMAP, is a pitch about self-sufficiency rather than about adopting supplied assets. THIS IS THE CELL TO REVISIT FIRST when the marketplace ships, and the timing is knowable: seed money was raised in May 2026 specifically to build it, so a lane-close or next-cycle pass should expect movement here before anywhere else on this record. No template library, agent gallery or prebuilt agent set is claimed anywhere in the material reachable.

F>P per the 1 September ruling, and product.php settles it beyond the Forbes block the ruling cited. THE ROUTING IS THE VENDOR'S, EXPLICITLY. The product page states that ReN3 INTELLIGENTLY ORCHESTRATES MULTIPLE AI MODELS, AUTOMATICALLY ROUTING EACH REQUEST BASED ON REASONING COMPLEXITY, LATENCY, ENTERPRISE POLICY, AND TOKEN COST, and closes the sentence WITHOUT MANUAL INTERVENTION. That last phrase decides the cell: the vendor is describing the absence of customer selection as the feature. Under the 10web ruling vendor-controlled routing is Partial, and the ruling's own test, client selection per task, is not met. THE ORIGINAL FULL RESTED ON A CLAIM THE VENDOR DOES NOT MAKE. The July basis said clients choose the best model per task, which came from an aggregator profile. What the vendor says is that it chooses, on the customer's behalf, against four named inputs. WHAT IS GENUINELY STRONG HERE AND EARNS THE PARTIAL COMFORTABLY. Enterprise policy is one of the four routing inputs, so an organisation shapes the routing envelope even though it does not pick per request. Locally hosted models including Llama, Mistral and Deepseek run entirely inside customer infrastructure, which is the deepest form of model control available to an air-gapped buyer: the question of which provider sees the data does not arise. Automatic provider failover keeps the service available regardless of external dependencies. THE COMMERCIAL POINT IS THE ONE TO CARRY, because it explains why routing is designed this way. The homepage reports a 67 PERCENT REDUCTION IN TOKEN SPEND, and the Forbes Asia 100 to Watch 2026 citation highlights the company for OPTIMISING COSTS THROUGH INTELLIGENT MODEL ROUTING. This is routing sold as a cost instrument rather than as flexibility, which is a coherent product decision and not a gap. WHAT WOULD MOVE THIS TO FULL: a documented per-agent or per-task model selection surface. plans.php and the product videos are where that would appear.

Stands at N, and the absence is now observed against the full site map rather than inferred from a failed retrieval. WHAT WAS ACTUALLY READ THIS PASS: the complete navigation and footer, which are Product, Solutions by industry and by use case, Plans, Partnership, Resources with product videos, newsletter and blog, and Company with about, core values, core team, newsroom and contact. There is no Developers entry, no API entry, no documentation entry and no docs subdomain anywhere in either the header or the footer. The homepage, product page and plans page carry no API, SDK, MCP or webhook mention. THAT IS A MATERIALLY DIFFERENT BASIS FROM THE JULY ONE, which recorded no public API documented and rested on a site nobody had reached. A vendor's complete published navigation is the right surface against which to conclude that a developer surface is not offered, and this one has seven top-level sections without one. THE ONE ADJACENT FACT, RECORDED AND NOT CREDITED: plans.php states that under Bring Your Own Cloud, LLM HOSTING CAN BE GPU/API-BASED. That is the customer pointing ReN3 at a model endpoint, which is the platform calling out. It is a deployment and model-hosting fact, credited on Dep and Model, and is the opposite direction from what this axis measures. THE ABSENCE IS COHERENT WITH THE SALES MOTION AND WORTH SAYING SO. Every route into this product is Book A Demo, Talk To Us or Enquire Now; pricing is annual subscription with concurrent-user site licensing and no self-serve tier; and the channel runs through Red Hat, Dell, NTT and Capgemini. This is a partner-led enterprise sale into government and banking, where integration is scoped in a statement of work rather than self-served from a developer portal. A published API would serve a buyer this company has not chosen. WHAT WOULD MOVE IT: an API or integration section on partnerships.php, or developer material behind the v1.ren3.ai application, neither of which was reached.