Product Roadmap

Langfuse Docker Compose, prompt_service.py, Prompt seeding script, Enrichment task migration, Suggestion + extraction migration, Prompt report CLI.

Theory strength analysis, Gap identification, Issue spotting, Counter-argument surfacing, Theory comparison.

PII redaction, Full self-verification.

Text message ingestion, Web capture, Email capture (OAuth), Email capture (intake forwarding), Celery Beat scheduling.

Client auth, Evidence upload, Incident log, Secure messaging, Evidence status, Client timeline.

Entity popover with cross-evidence counts, "Next mention" navigation, Mobile jump-to-mention, Fact source snippet highlight in extracted text, Inline fact approve/dismiss on evidence detail.

Timeline endpoint, Horizontal timeline, Event markers, Detail disclosure, Linked views, Filter bar.

Issues, Issue-fact linking, Issue-evidence linking, Claims & issues management, Issue-fact linking UX, Issue-evidence linking UX.

Report infrastructure, Case Timeline, Annotated Timeline, Evidence Index, Entity Map, Evidence Ledger.

Per-case cost tracking, Per-case monthly cap, Firm monthly cap, AI operations log, Rate card pricing engine, Attorney management.

PrivateLink, VPC, ZDR. Swap Anthropic API for Bedrock before onboarding paying customers. S3 hardening: upgrade to SSE-KMS (customer-managed keys), enable bucket versioning, MFA delete protection, cross-region replication. DB scaling: range-partition events table by `created_at` (monthly). Quotas: per-case evidence item count limit (configurable per firm via `firm.settings`). Case-level RLS: add `case_id` RLS policies on KG tables (`entities`, `facts`, `relationships`, `entity_evidence`, `fact_evidence`) to enforce case isolation at the database layer — defense-in-depth behind application-layer participant checks. Required before M17 multi-agent case teams to prevent cross-case context leakage.

Client Q&A with agentic retrieval. Attorney-facing mode: conversational querying over the full evidence vault from the attorney dashboard — real-time fact-checking, contradiction detection, and argument stress-testing during active case work. Client mode bounded by client permissions; attorney mode has full case access including work product.

Synthesis, comparison, detection

Manifest reasoning, model routing

API keys UI, sessions, automation rules

Intake, Research, Drafting, Client-Facing agents. Agent instruction layer accepts firm-level and case-type-level instruction profiles (see [Instruction Profiles](features.md#instruction-profiles)). Architecture for profile resolution ships in M17; profile authoring UI ships in M22+. Requires case-level RLS from M12 to prevent agent cross-case memory leakage.

Transcript viewer, citations, designations

Automated monitoring, audio/video, screen recording

Saved searches, alerts, relationship queries

Entity Dossier, Case Brief, Deposition Prep, etc.

Value comparison, client-facing export, approval gates