How to run your practice on CaseOps.

CaseOps is a workspace for running a legal practice end to end — matters, documents, drafting, hearings, contracts, outside counsel, billing, and the audit trail that ties them together. You land on the home dashboard after sign in; everything else is one keystroke away from the command bar (⌘ K or Ctrl K).

1. 1
   Create a workspace. Visit /sign-in, pick New workspace, and enter your firm or company name. You become the first admin.
2. 2
   Invite colleagues. From Admin → Members, send invitations with a role — Partner, Associate, General Counsel, Legal Ops or Reviewer.
3. 3
   Verify the brand. Add your firm logo and a default contact email so draft dispatches and invoices carry them.
4. 4
   Open your first matter. Skip to section 3 if you are ready, or continue through sections 2 and 3 together for context.

Every workspace is an isolated tenant. Data never crosses tenant boundaries — the query layer, storage layer and audit trail all filter by tenant id. Users inside a workspace see what their role permits, and matter-level ethical walls can further restrict access below that.

Roles at a glance

• Admin. Full access. Manages members, teams, billing settings, AI policy and audit exports.
• Partner. Owns matters; approves drafts; signs off recommendations; sees team-wide portfolios.
• Associate. Opens and works matters; produces drafts; compiles hearing packs; requests approvals.
• General Counsel. Portfolio view; intake routing; contract and obligation oversight; outside-counsel spend.
• Legal Ops. Intake, SLAs, reports; no draft approvals.
• Reviewer (outside). Scoped, time-bound access to a specific matter; read-only by default.

A new matter takes about a minute to open and is the anchor for every other workflow. The fields below are the minimum that make downstream drafting, hearing prep and recommendations useful — a thinner record will still work but will lean on fact placeholders.

1. 1
   Go to Matters → New matter. Give it a clear plain-English title (e.g. Arbitral award challenge — Acme v. Kappa).
2. 2
   Pick a type (litigation, arbitration, advisory, contract, regulatory) and a forum (court, tribunal, internal).
3. 3
   Add parties on both sides. Each party row carries its counsel, contact and role.
4. 4
   Set the stage (pleadings, evidence, arguments, post-judgment). Stages drive deadline templates and surface only relevant actions in the cockpit.
5. 5
   Attach the first batch of documents. See section 4 for how indexing and OCR work.

The matter cockpit

Opening any matter puts you in the cockpit. The top bar shows parties, forum and the next hearing. Tabs run across: Overview, Documents, Drafts, Hearings, Research, Contracts, Billing, Team, Audit. Everything that happens here is recorded in the audit trail with the user, timestamp and before and after state.

Every document uploaded to a matter is queued for extraction and embedding so you can search it semantically alongside the CaseOps public corpus. Expect PDFs, DOCX, emails (EML) and scanned pages. Scanned pages run through OCR before indexing.

• Upload. Drag and drop onto the Documents tab, or paste a file list. Files above 50 MB are rejected at the edge.
• Status pills. Queued → Indexed. OCR pending appears on scans and typically clears in a few minutes.
• Retry. A failed extraction shows a Retry button. Re-indexing a document does not duplicate chunks.
• Private by matter. Documents inherit the matter's ethical wall. They are never pooled across tenants or used for cross-tenant training.

The Drafting Studio produces a first draft from the matter's own record — parties, stage, documents, focus note — grounded in statutes and judgments retrieved from the CaseOps corpus and your internal precedents. Every inline citation resolves to a named authority. Every fact gap renders as a placeholder the reviewer fills in, not as a fabricated number.

1. 1
   In a matter, open Drafts → New draft and pick a template (bail application, §34 petition, reply to summons, quashing, etc.).
2. 2
   Add a focus note — one or two lines of what this draft must argue. This is the single most load-bearing field.
3. 3
   Press Generate. The first pass finishes in 30–90s; the draft opens with inline citation pills and a grounding panel on the right.
4. 4
   Review for: fact placeholders to resolve, citations to verify, and statute attribution. The reviewer findings block at the foot of the draft calls these out.
5. 5
   Request approval from a partner. Approval is recorded in the audit trail alongside the draft version.

The Hearings tab pulls the next listings for this matter. Open the next hearing and press Compile pack. CaseOps stitches a pack in under a minute, from the matter record and the authority corpus:

• A chronology built from the matter's documents and activity — no manual entry.
• The last order and its operative portion extracted and pinned.
• A short oral points list — the arguments you actually want to make, keyed to the matter record.
• Source-backed bench context with sample size, linked sources, and limitation notes where the corpus supports it.
• The source list — every piece of content in the pack is traceable back to a matter document or a named authority.

Litigation Intelligence is the matter-level workspace for source-backed preparation and review. It pulls together proceeding sheets, affidavits, mock-hearing sessions, predictive context, legal-source readiness, a matter knowledge graph, and transcript-first coaching without changing the rule that a lawyer reviews substantive output.

• Proceeding Sheet Intelligence. Court orders and order sheets are parsed from raw order text only. Next hearing dates, compliance directions, affidavit deadlines, and generated tasks keep the source order, snippet, confidence, and review-required status.
• Affidavit Intelligence. Mark a document as an affidavit, chief affidavit, or counter-affidavit to extract key statements, dates, figures, entities, annexures, gaps, contradictions, and source-grounded cross-examination questions.
• Mock Hearing and Coach. The simulator is typed-text only and uses LI affidavit question banks. The coach requires a session acknowledgement and scores observable preparation markers such as whether the question was answered, whether a source reference was used, and whether unsupported assertions were added.
• Predictive and bench context. Predictive Intelligence shows observed historical patterns only when indexed source evidence exists. Supported signals display sample size, confidence band, source links, snapshot references, and limitation notes.
• Review queue and knowledge graph. The Litigation Intelligence review page groups pending review items and lets permitted users accept, reject, mark reviewed, or edit notes. The knowledge graph materializes matter-scoped nodes and edges from source-backed LI records with bounded snippets.

When you generate an appeal_memorandumdraft for a matter that has an upcoming listing, CaseOps doesn't just pull authorities from the court at large — it pulls authorities authored by the specific bench scheduled to hear the appeal and prefers ones that align with the matter's practice area. The advocate-bias selection is editorial: the system surfaces the citations that support your grounds. Adverse-authority duties to the court remain yours.

• Career timeline on every judge profile — /app/courts/judges/{id} shows every court the judge has served on, with source-attributed evidence.
• Bench resolveron the matter hearings tab — "Justice X & Justice Y" renders as clickable links to each judge's profile.
• Appeal Strength panel on the appeal stepper flags per-ground citation coverage and weak-evidence paths.

Visit /app/statutes to browse the structured catalog of central Indian Acts. v1 ships with 7 acts and 91 sections, each with a source link to indiacode.nic.in — the Government of India's official Acts repository (public domain).

• Attach to a matter. The Statutes sub-tab on the matter cockpit lets you mark which sections this matter cites, opposes, or holds in context.
• Fed into drafting. When you generate an appeal-memorandum draft, the prompt receives the bare text of each attached section. The LLM is instructed to quote verbatim instead of paraphrasing.
• BNSS vs BNS unambiguous.The structured reference makes the act explicit, so "Section 483 BNSS" (bail) is never confused with "Section 483 BNS".

The Research workspace (⌘ K then type research, or visit /app/research) is the place to run open-ended queries against the public corpus without tying the query to a specific draft. It is a search tool, not a chatbot.

• Query. Natural language works best — phrase the issue, not a keyword. Example: triple test for anticipatory bail under BNSS s.482.
• Results. Judgments return with a short extract, court and date metadata, and a link to open the full PDF. Cosine distance is surfaced so you can judge strength at a glance.
• Annotations. Flag or shortlist an authority for the current matter. Annotations are tenant-private and travel with the matter, not the user.

Contracts are a first-class matter type. You can open a matter that is a single contract review, or attach a contract to a litigation matter. The Contracts tab shows clause extractions, obligations, redlines and version lineage.

• Upload. Drop the contract (.docx or .pdf). CaseOps extracts parties, effective dates, key covenants and payment terms.
• Playbook compare. Pick a playbook; the system flags clauses that deviate from it and suggests an edit per deviation.
• Obligations. Payment, reporting, consent and termination obligations lift into the Obligations list with due dates and owners.
• Redlines. Track changes export cleanly to Word with a single summary of what changed and why.

CaseOps produces explainable recommendations — forum choice, supporting authorities, next best action — with rationale, assumptions, missing facts and a confidence label. Recommendations must show the source trail and limitation note before anyone accepts them.

1. Rationale. Two to four sentences, grounded in named authorities and the matter record.
2. Assumptions. Facts the system took as given. Wrong assumptions are your signal to reject the recommendation.
3. Missing facts. Fields the system flagged as absent or too thin. Filling these in and re-running sharpens the output.
4. Confidence. High / Medium / Low. Low-confidence recommendations are deliberately surfaced, not hidden.

General Counsel teams run Outside counsel from the top-level nav. The directory carries rate cards, historical outcomes, speciality tags and conflict flags. A matter can brief one or many firms; spend and realisation roll up to the portfolio view.

• Brief. From a matter, pick a counsel; CaseOps issues a brief packet that respects the matter's ethical wall.
• Budget. Assign a budget with alert thresholds. Invoices above the cap require a partner override.
• Realisation. The counsel dashboard shows billed, collected and aging, with the portfolio roll-up on the home dashboard.

CaseOps produces invoices from tracked time and matter activity. Pine Labs is wired as the default payment rail — every invoice goes out with a payment link, and settlement writes back as paid automatically.

• Time entries. Log from the matter or the global Time tab. Filters for user, matter, client and date range.
• Invoices. Draft → send → paid. Drafts pull WIP time and matter-level expenses; sends attach a PDF and the payment link.
• Recoveries. Partial payments, holds and write-offs are first-class, not free-text notes. Every state change is on the audit trail.
• GST. India GST is handled on the invoice line-item level; a monthly report exports to the format your accountant expects.

Admins run the workspace from /app/admin. The six important subsections are Members, Teams, AI policy, Audit export, Ethical walls and Billing (workspace billing, not client billing).

• Members. Invite, change role, deactivate. Deactivation revokes access immediately; the user's prior actions stay attributed in the audit log.
• Teams. Matter visibility can be scoped to a team. Useful for practice groups or conflict segregation short of a full ethical wall.
• AI policy. Cap the providers, models and context shapes the workspace is allowed to use. AI actions outside the policy are blocked before any token is spent.
• Audit export. JSONL or CSV for any date range. Large exports run as background jobs; download when complete.
• Ethical walls. Add, edit, dissolve. Every change is itself an audited event.

Security is not a tab — it is the way every other tab is built. The short version:

• Tenant isolation at the query and storage layer, not only the application layer.
• Matter-level ethical walls override role access. A partner outside a wall sees no traces of the walled matter.
• Audit on every sensitive action — create, read, update, delete, export, AI run, recommendation accept, payment state change.
• No cross-tenant training. Your documents and matter activity are not pooled into model training without an explicit opt-in.
• Encryption in transit and at rest, signed URLs for document downloads, and short-lived session tokens.
• Hardened browser headers — CSP, HSTS, X-Frame DENY, strict Permissions-Policy — reduce client-side exposure.

A document won't index

Check the status pill in Documents. Most failures are OCR timeouts on very large scans — press Retry. Persistent failures usually mean the PDF has no extractable text and no OCR layer; convert to DOCX or re-scan at a higher DPI.

Citations in my draft look wrong

Use the grounding panel on the right side of the Drafting Studio — every inline citation has a source. If an authority is wrong for the point, open it, remove it from the shortlist, and regenerate. The reviewer findings block at the foot of the draft also flags likely mismatches.

Research returned 0 results

Two common causes. The query is too generic (bail) and the system refused to narrow arbitrarily — rewrite with the issue and the statute. Or the corpus does not cover that jurisdiction and year — the Home dashboard shows corpus coverage by court and year.

A colleague cannot see a matter I opened

Ethical wall or team scope. Open the matter, check Team, and either widen the team or dissolve the wall. Admins can do this; partners inside the wall can too.

Matter graph

The connected record of everything that belongs to one matter — parties, documents, drafts, hearings, invoices, activity, audit.

Cockpit

The single-page view of a matter. Every tab inside it is a lens on the same graph.

Corpus

The public pool of statutes, judgments and regulatory material CaseOps retrieves against. Tenant documents are separate and tenant-private.

Ethical wall

A matter-level access control that restricts a matter to named users, overriding broader role permissions.

Grounding

The link between an AI output and the source it came from (a judgment, statute, or matter document). A CaseOps output without a grounding is a bug.

Recall@10 / MRR

Retrieval quality metrics. Recall@10 is the fraction of queries whose correct answer appears in the top-10 results. MRR (mean reciprocal rank) averages 1/rank across queries — higher means the correct hit sits closer to the top.

Hearing pack

The bundle a lawyer takes into court — chronology, last order, oral points, bench brief, source list — compiled from the matter record and authority corpus.

Playbook

A named set of preferred contract positions. CaseOps compares any inbound contract against the playbook and surfaces deviations.