// CHANGELOG
What shipped, when it shipped, and what changed for the people using it. Entries are grouped by release, newest first, and describe behaviour you can observe in the product.
v1.3.0 Alpha is a pre-release. It is feature complete for the workflows documented below and is suitable for evaluation and pilot use. It is not yet recommended as the system of record for a statutory filing without review. Interfaces and data models will continue to change ahead of beta, and breaking changes will be announced here before they ship.
Read the v1.3.0 Alpha notesFrontier Sustainability Intelligence & Statutory Surrender Defense
October 2026
Pre-release development window: October 2026
ZeroCarbon
ZeroCarbon v1.3.0 Alpha introduces frontier cognitive and operational capabilities into ZeroCarbon Cowork, establishing audit-grade mathematical rigor and executive statutory defense for global decarbonization operations. This release integrates deep industrial utility and multimodal ledger reasoning (kVAh billing, Time-of-Day tariff slots, captive generation reconciliation), autonomous multi-step goal deciders with dynamic replanning, executive statutory proof for EU CBAM Definitive Regime Article 9 carbon price surrender deductions and SEBI BRSR Core Principle 6 reasonable assurance, and sub-second deterministic calculation via @zerocarbon/emissions-core. All capabilities adhere strictly to international public standards without exposing proprietary company methods.
Cowork now autonomously decomposes complex decarbonization directives into governed multi-agent execution sequences across discovery, deterministic classification, anomaly detection, gap analysis, and executive artifact compilation. Operates with dynamic re-planning budgets and proactive cross-verification between ERP financial invoices and smart meter telemetry.
Extends document and ERP parsing to handle complex industrial utility structures: distinguishing active power (kWh) from apparent billing units (kVAh), extracting Time-of-Day (TOD) peak/off-peak tariff zones, recording Maximum Demand (MD) contract exceedances, reconciling captive diesel generator fuel logs with standard density metrics (0.84 kg/L), and handling solar net-metering and Green Open Access wheeling attributes.
Audit-grade statutory defense anchored strictly in public regulations: EU CBAM Regulation (EU) 2023/956 Definitive Regime (formulating legal and quantitative proof for Article 9 carbon price deductions to offset domestic CCTS or carbon excise payments against EU certificate liabilities, and Article 35 installation-level primary benchmark defense), SEBI BRSR Core Circular 2023/122 Principle 6 reasonable assurance for Top 250 value chains, and WBCSD PACT Methodology Version 3 (v3.0) 5-dimensional Data Quality Indicator (DQI) ratings.
High-throughput line extraction and batch calculation powered strictly by the verified in-process @zerocarbon/emissions-core engine (CEA India v21, DEFRA 2026, EPA eGRID 2026, IPCC AR6 GWP100). Eliminates LLM math approximation and enforces cryptographic SHA-256 evidence provenance.
Interactive AI diagnosis for operational sustainability queries, synthesizing public statutory guidelines, root-cause analyses, and instant 1-click pre-filled prompt handoffs directly into the Cowork agent workspace.
Official high-fidelity vector provider identities via Logo.dev CDN with fallback resilience, paired with full theme-adaptive contrast and ambient lighting across light and dark modes.
Company Brain
October 2026
Development window: late September to October 2026
ZeroCarbon
ZeroCarbon Cowork now learns how your organization works, and the workspace around it has been rebuilt to match. The Company Brain turns the know-how scattered across documents, conversations, approvals and settings into reviewed, executable skills that Cowork follows. Reports are ready to share under your own brand, GLEC and PACT calculations are tested against the frameworks' own worked examples, and the dashboard has been redesigned around analytics you can act on. This remains a pre-release: interfaces and data models will continue to change ahead of beta.
Cowork reads the places your organization's know-how already lives - uploaded documents, what your team tells Cowork, approval decisions, feedback and workspace settings - and extracts the policies, procedures, decision rules and owners specific to you. Every item quotes the source it came from, and anything that cannot be traced to a source is discarded. Each process is compiled into a versioned skill with steps, decision rules, required approvals and escalation owners, and Cowork only runs a skill once an owner or admin has approved it. When the underlying knowledge changes, the updated skill waits for review while Cowork continues to use the approved version. Teams can also teach Cowork directly, connect other systems through the API, and export approved skills in the standard Agent Skills format.
zc, the ZeroCarbon mascot, now joins you across the product. On a new conversation it briefs you on what needs your attention: tasks waiting for an answer, changes waiting for approval, background work in progress and how recent your data is. During a conversation it narrates the run as it happens - which specialist is working, which step it is on and how long it has taken - and tells you when an approval is needed or the work is done. On other pages it reminds you, once, when a background task needs you. Everything zc says is taken from your workspace's live state.
The overview now opens with a personal welcome and setup guide, followed by metric tiles for total emissions, scope mix, data confidence and compliance readiness, each with real period-over-period change. An interactive trend explorer shows monthly emissions by scope, with a drill-down on any month. New panels show where your emissions come from and how much of the footprint rests on measured, verified and documented data. Ask Cowork bars on the dashboard and key pages open Cowork with your question already written, and every company page now shares one consistent design.
The GHG inventory report is rebuilt directly from your ledger records on a flowing layout that keeps tables and sections intact across pages. Reports carry your company logo and authorised signatory alongside the ZeroCarbon mark, and draft stamps are replaced by a clear status and a count of outstanding items. CBAM communications include charts of embedded emissions.
Freight calculations use the GLEC Framework v3.2 Module 1 regional fuel tables and Module 4 methods, and product footprints report PACT Methodology v3.0 data reliability through the primary data share and data quality ratings. Both are verified against the worked examples published in the frameworks themselves, and both are available to Cowork as tools.
Independent specialists now run at the same time, while anything that needs your approval still runs one step at a time. Ask why a figure is what it is and Cowork walks the stored evidence - the records, factors and approvals behind it - instead of recalculating. Every emissions figure Cowork states comes from the calculation engine or your ledger, never from the model alone.
Automations can now run when data changes - for example when a record is corrected or a factor is retired - as well as on a schedule. A Tally bridge brings in the Day Book from TallyPrime, with every entry reviewed before it is recorded. Suppliers are matched by GSTIN first, and a name never overrides a conflicting identifier.
Consent given at sign-up, during workspace setup and through the cookie banner is recorded with the exact notice shown, the date and time, and where it was given. Owners and admins can review the full consent history in Data control, withdraw consent in a single step and give it again, and erasure is offered as a separate, explicit choice. Cookie preferences can now be managed individually.
ZeroCarbon Cowork
September 2026
Development window: August to September 2026
ZeroCarbon
A major milestone that transforms our legacy passive AI document ingestion into ZeroCarbon Cowork: an enterprise-grade, autonomous carbon accounting agent workspace. Where AI Ingestion was previously a passive OCR scanner with a static review queue, ZeroCarbon Cowork (/company/agent) is an active, conversational, multi-agent co-pilot that actively reasons across operational ledgers, utility feeds, supplier declarations, and ESG compliance frameworks. Built on a LangGraph ReAct orchestration runtime with strict human-in-the-loop interruption safety, transparent step-by-step reasoning traces, real code and factor diff viewers, interactive Generative UI cards, and bidirectional enterprise data connectors.
The legacy /company/ai-ingestion interface has been retired and fully reimagined as ZeroCarbon Cowork at /company/agent. Rather than waiting for users to upload isolated documents and manually inspect tabular extractions, Cowork acts as a dedicated ESG partner. It proactively discovers anomalies, suggests GHG classifications across Scope 1, 2, and 3, runs conversational audits on carbon intensity, and executes complex multi-step analysis across millions of operational records in seconds.
ZeroCarbon Cowork is powered by an asynchronous multi-agent supervisor architecture built on LangGraph. Incoming requests are dynamically routed across specialized sub-agents: Document Extraction Agent (VLM OCR and multimodal document intelligence), Emissions Verification Engine (factor resolution and GHG math against verified factor databases), Compliance & Governance Agent (CBAM, BRSR Core, CSRD, and SEC climate checks), and Supply Chain Scope 3 Agent (PACT Pathfinder counterparty resolution). Each agent operates with discrete tool calling capabilities, stateful checkpointing, and execution isolation.
The chat feed is no longer limited to static markdown text. Through our zero-latency Generative UI renderer, Cowork streams interactive, clickable enterprise UI widgets directly into the conversation. Users can interact with live Scope 1-3 intensity breakdown charts, dynamic factor comparison tables, supplier decarbonization ranking scorecards, and clickable CBAM liability projection cards without navigating away from the workspace.
Every agent action is completely transparent. The new unboxed ThinkingState component displays animated 3x3 pixel dot indicators and chronological thought processes showing AST checks, database queries, and factor resolutions. Embedded TerminalCommand viewers stream mock and live bash execution results with exit codes, while visual FileDiff inspectors render addition and deletion gutters (+3 / -1) whenever the agent proposes a patch or adjustment to emission configurations.
To guarantee strict auditability and prevent unauthorized ledger mutations, Cowork enforces a zero-trust propose-then-approve architecture. When an agent identifies a necessary ledger write, it halts execution via a LangGraph interrupt frame, packaging the proposed changes into an interactive ApprovalCard. Only when a human operator verifies and clicks 'Approve' is the transaction cryptographically recomputed and idempotently committed to the certified ledger with SHA-256 audit chaining.
Hallucinated calculations are eliminated. Every numeric quantity produced by ZeroCarbon Cowork is strictly resolved against our single shared factor registry. Calculations output real-time factor citation badges linking directly to their authoritative source (such as CEA CO2 Baseline v20, GLEC v3.2, DEFRA 2025, or EPA eGRID), carrying the exact methodology, GWP version, and verification timestamp into the record.
Cowork features a seamless dual-mode interface: a conversational agent chat on the left and an expandable visual Workflow Canvas on the right. Operators can visually inspect n8n-style workflow pipelines, monitor active node execution states, debug pipeline branch conditions, and chain automated tasks ranging from automated monthly invoice reconciliation to bulk supplier data requests.
Cowork natively connects to enterprise data sources through a dedicated Connectors Panel. Operators can configure automated webhooks, pull live ERP purchase orders from SAP and NetSuite, ingest SCADA utility meter feeds, and connect third-party AI agents via the Model Context Protocol (MCP).
An integrated AIContextMeter continuously tracks conversation token capacity, active context usage percentages, and provider usage metadata in real time. Combined with our progressive token reconciler, assistant responses stream smoothly at sub-20ms latency with live follow-up queueing (A7) that stashes user prompts typed mid-stream and executes them cleanly upon turn completion.
The Cowork workspace introduces a bespoke WebGPU ambient fluid background that gently shifts based on agent activity, creating a state-of-the-art visual experience. The header dynamically personalizes time-aware professional greetings adapted to the user's organization, while the layout fluidly adapts across desktop and mobile screens.
Platform Re-Architecture
September 2026
Development window: July to September 2026
ZeroCarbon
The largest release in the platform's history, and the first to carry the ZeroCarbon name. Almost every subsystem was rebuilt around three requirements: every reported figure must be traceable to a named emission factor and its source, every write must be idempotent and recorded, and every read must be scoped to the organization that owns the data. This is a pre-beta Alpha. It is feature complete for the workflows described below, but APIs and data models will continue to change ahead of beta.
The emissions calculator now lives in a single shared package that both the web application and the backend services compile against. Four separately maintained emission factor tables had drifted apart; they are consolidated into one resolver. The Indian grid factor, which previously appeared as four different values across the codebase, resolves to a single canonical figure of 0.708 kgCO2e/kWh sourced from the CEA CO2 Baseline Database. A quantity calculated in the ingestion pipeline, in a report, and through the API now returns the same result because it is the same code path.
Factors resolve through a database backed registry with a defined precedence: an organization's own registered factor, then the shared global registry, then a verified built in floor. An organization on a captive power purchase agreement can register its contracted grid factor and have every calculation and scenario use that figure instead of the national default. Each resolution carries its origin, match specificity, source, methodology, GWP version and validity window into the audit trail, so a filed report can state where each number came from.
Freight emissions are calculated per transport chain element rather than as a single blended figure. Tonne-kilometres are derived per leg, Well-to-Wheel and Tank-to-Wheel intensities are carried separately, factors are selected per fuel and region rather than from one diesel constant, and load factors are bounded and account for empty running. Data quality is scored across five dimensions from the inputs actually supplied. A shipment missing a distance now returns an explicit error rather than substituting an assumed value, and every result carries a SHA-256 calculation hash for verification.
Product carbon footprints can be produced from verified ledger records and exchanged with customers over the WBCSD Pathfinder network interface. Emissions are attributed through three tiers, being direct records, production batch records, and allocated facility overhead, with the allocation basis recorded on each component so a buyer can trace any figure back to its evidence. Access is granted per counterparty and can be revoked. Footprints are versioned and superseded rather than edited, so a footprint a customer has already cited remains retrievable.
Disclosure reports, SEBI BRSR Core reports and EU CBAM reports now render as formatted PDF documents suitable for filing, with BRSR XBRL and CBAM CSV and XML output for machine submission. Every generated report is anchored as an immutable SHA-256 hashed record at the moment of issue, so a report distributed to an auditor or regulator can later be shown to be unaltered.
Invoices, utility bills and shipping documents are read by a vision-language model tier for scanned images alongside a separate extraction tier for documents that carry real text. Where a figure cannot be read with sufficient confidence the pipeline reports that it could not be read. It does not emit an estimated quantity, because an unreliable number that looks like a measurement is more damaging in an audited report than a gap.
The assistant layer runs against the production calculation engine rather than simulated arithmetic, and operates on a propose-then-approve model. An agent drafts a record, a person reviews it, and only on approval is the figure recomputed and committed to the ledger. Approvals are persisted at the point of interruption and enforced as idempotent at the database level, so an approval replayed twice cannot produce two records.
Records identified as incorrect are quarantined and restated rather than overwritten, so the original figure remains available to anyone reconciling a previously issued report. The exclusion rule is enforced in the database client itself, which means every total, grouping and count honours it without relying on each query to remember to apply it.
A new public site built on the Clay design system, with technical landing pages for EU CBAM, CSRD ESRS E1, GLEC v3.2, PACT v3.0, SEBI BRSR Core and the GHG Protocol Scope 1-3 API, alongside About, Team, How It Works, Impact and Careers. Includes structured data, a sitemap, and an llms.txt description for AI crawlers.
This changelog, and a public status page whose indicators are measured on every page load rather than published from a static file. Uptime history accrues from recorded checks, so the page reports a percentage only once it has the measurements to support one.
Hardening and Distribution
July 2026
CarbonLedger era
The final CarbonLedger release before the platform was renamed and re-architected as ZeroCarbon.
AI Ingestion and MCP
June 2026
CarbonLedger era
The platform learned to read documents, and became addressable by AI tooling through the Model Context Protocol.
Supplier Network
May 2026
CarbonLedger era
Suppliers gained a route to submit primary data directly, rather than being estimated from spend.
Compliance Surface
February 2026
CarbonLedger era
Foundations
January 2026
CarbonLedger era
Phase 1
29 December 2025
CarbonLedger era
The first product milestone. Organizations could sign in, request a demo, and begin recording emissions.
Genesis
12 December 2025
CarbonLedger era
The first commit, and the first deployment.
Releases are dated to the month the work shipped. Where a release covers a longer development window, that window is stated on the entry. Security items describe the protection now in place rather than the detail of what preceded it. For current service health, see the status page.