Agentic AI Platform Change Log

Workflow orchestration

Legal Work Orchestration organizes a matter in Harvey into phases and tasks, each carrying its scope, inputs and reviewer, so agents can execute defined tasks and lawyers review the output at designated gates. Harvey notifies the reviewer when an agent finishes, moves to the next step on approval, and lets lawyers close or start tasks by replying to email, while reusable matter templates are in early access.

Harvey moves from assisting prompt by prompt to agents running defined steps of a matter under explicit approval gates, the structure firms need before delegating more work to agents.

MCP / tool calling / API

A new Connector Library in Harvey is a catalog of admin enabled connectors that are either native integrations or vendor run MCP servers its agents call as tools. Harvey says each MCP tool is security reviewed first, every call passes a policy engine that fails closed, and admins can switch off a whole connector or a single tool.

Firms can let Harvey agents work across outside systems while keeping per tool control and an off switch, the safeguard security teams ask for before agents touch client data.

Integrations

Harvey added agentic search for connected SharePoint, OneDrive, and Google Drive content. Users can have Harvey find relevant documents directly instead of locating and selecting the files themselves.

Legal teams evaluating access to institutional knowledge can assess whether Harvey retrieves the right material from their existing document systems with less manual preparation.

Agent capability

Harvey lets users hand any Review Table action to the Harvey Agent in plain language, including reclassifying a column, importing metadata as new columns, bulk editing cells and routing flagged rows to a reviewer.

Review tables change as a matter changes. Ask how the agent's edits are logged, and whether rerunning a column keeps cells a reviewer already verified.

Agent capability

Harvey's September update lets users attach up to five knowledge sources to a single query and adds more than 150 new legal research sources.

Multi source queries let a lawyer ask one question across a data room, a knowledge base and research at once. Check that answers cite which source each point came from.

Integrations

Harvey released a suite of platform updates including a native iManage search integration, over 130 new legal research sources, and client matter governance enhancements. The Harvey for Word add-in also gained the ability to apply suggestions from historical threads.

Firms using iManage can now search their document management system directly from Harvey. The expanded research sources and matter governance tools improve the platform's utility for enterprise legal teams.

Agent capability

Harvey introduced Tenet, its first post-trained open-weight AI model designed specifically for long-horizon legal tasks. Built on the Kimi K3 base model and developed with Fireworks research, Tenet leverages synthetic data and expert human inputs to improve performance on complex workflows like regulatory analysis and trial preparation.

Provides law firms with a highly specialized, domain-specific model that outperforms general-purpose frontier models on complex legal reasoning and contract analysis benchmarks.

Agent capability

Harvey released several platform updates, including new agent actions that allow users to bulk edit, organize, and import metadata directly into review tables. The update also introduced enhanced multi-source queries, enabling users to attach up to five knowledge sources in a single prompt, and added alternate suggestions to the Harvey for Word add-in.

Enhances user productivity by streamlining bulk document review tasks and improving the depth of legal research through simultaneous multi-source querying.

Observability / auditability

Harvey released a suite of updates highlighted by Command Center, a new analytics workspace for administrators to track adoption and query workspace data. The release also introduces Mistral Medium 3.5 as a supported model, sets DeepL as the default translation provider for US and EU workspaces, and adds Mexican legal content from Tirant lo Blanch.

Legal operations leaders gain visibility into platform usage and ROI through the new admin dashboard. Practitioners benefit from expanded model flexibility, dedicated translation engines, and broader jurisdictional coverage for Mexican legal research.

Security / enterprise

Harvey released its generally available integration with Intapp Walls for AI. The update allows law firms to sync their existing ethical walls and information barriers directly into Harvey. It automatically enforces matter-level restrictions and conflict-of-interest policies across Harvey Assistant, Vault, and Shared Spaces.

Confidentiality risk is a primary barrier to firm-wide AI deployments. By inheriting established Intapp policies rather than relying on manual configuration or attorney discipline, law firms can confidently scale Harvey usage while maintaining strict compliance controls.

Memory / state

Harvey shipped Org Context, which lets firms give the platform standing knowledge of their organization (practice structure, precedents, preferences) so outputs reflect the firm rather than a generic baseline, alongside Internal Spaces for firm-internal collaboration and the ability to attach documents directly to review tables.

Standing organizational context reduces the per-matter setup work of re-explaining firm conventions, and Internal Spaces keeps collaborative legal work inside the governed platform instead of scattering across email and shared drives.

Agent capability

Harvey made Claude Sonnet 5 available across its platform, reporting 91.3% on BigLaw Bench, its benchmark of real legal tasks. The upgrade applies to Assistant, Vault review, and Harvey's agent workflows.

A measurable jump on a legal-specific benchmark, not just a model swap. For legal buyers, published task-level scores give a concrete basis to compare platform quality as the underlying models change.