Create a full PRD for OpenBrain

[SorraTheOrc]

Headline: 6-month local-first PRD for OpenBrain focused on ingestion, search, and scheduled agents.

Problem statement
OpenBrain lacks a consolidated Product Requirements Document (PRD) describing a clear product vision, target users, MVP scope, success metrics, constraints, and a 6‑month roadmap focused on a local‑first CLI experience.

Users

• Developers & engineers: need clear user stories, acceptance criteria and implementation guidance to build features and tests.
• Agent operators: need scheduled retrievals/briefings and local agent workflows to automate tasks.
• Contributors: need runnable local dev environment, tests, and plugin interfaces for ingestion/search.

Example user stories

• As a developer, I can ingest files/URLs via the CLI and generate embeddings stored locally so I can build searchable memories offline.
• As an agent operator, I can schedule retrievals and briefings that run locally and produce concise summaries for automated workflows.
• As a contributor, I can run the project locally (Linux/macOS), run tests, and extend ingestion/search plugins with documented interfaces.

Acceptance criteria (per user story)

• Ingest: a documented CLI ingest flow exists; embeddings are generated and persisted locally; a small test corpus can be ingested and later retrieved by search.
• Scheduled retrievals/briefings: a scheduler interface (CLI or config) can run retrievals locally and produce concise summaries; automated test verifies a scheduled run completes and writes output.
• Contributor/onboarding: repository contains clear developer setup instructions and tests that run locally on Linux/macOS (or WSL); a sample plugin template documents the ingestion/search extension points.

Success criteria

• Deliver 10 core stories with acceptance tests passing (≥90%).
• Search relevance: precision@5 ≥ 0.8 on a small human-rated benchmark.
• Feature delivery milestones: complete the MVP feature set for local ingestion, embeddings, search, and scheduled retrievals within the 6‑month roadmap.

Constraints

• Local‑first CLI-only implementation for the 6‑month plan (SQLite + sqlite-vec, CPU-friendly models by default).
  -- Network access is allowed for features that benefit from it (for example: optional remote embedding providers, connectors, or telemetry). Core functionality MUST NOT rely on proprietary network services — any networked provider or hosted model must be optional, pluggable, and have a documented local fallback that preserves core capabilities and privacy.
  -- Support Linux + macOS; Windows support is not required — the project is expected to run under WSL on Windows where needed.
  -- Optional GPU acceleration if available.

Risks & assumptions

• Risks:
  • Reliance on hosted/proprietary models without local fallbacks could break privacy and availability — mitigation: require pluggable providers and document local fallbacks.
  • Scheduler/agent reliability (local execution) may suffer across environments — mitigation: prototype scheduler early and add integration tests.
  • Search quality may vary with CPU-only models — mitigation: include a small human-rated benchmark and evaluate tradeoffs; allow optional higher-quality remote providers.
  • Scope creep: adding many integrations/features in the PRD risks bloating the MVP — mitigation: record additional features as separate work items linked to this epic.
• Assumptions:
  • Local-first deployment using SQLite + sqlite-vec is feasible for MVP workloads.
  • Contributors will use Linux/macOS or WSL for development and testing.

Existing state

• README describes OpenBrain as a CLI-first memory system and references local-first implementations (SQLite + sqlite-vec) and external projects (OB1, SourceBase) for inspiration.
• Current work items include brainstorming, memory compartmentalization, and integrations with OpenBrain-related recipes (see related work below).

Desired change

• Produce a PRD that defines product vision, 6‑month roadmap, prioritized feature list (ingestion, embeddings, search, scheduled agents), measurable success metrics, constraints, and clear acceptance criteria for each MVP story.

Deliverables

• docs/PRD.md — the full PRD in markdown including user stories, acceptance criteria, constraints, roadmap, and success metrics.
• Acceptance tests and a small example dataset for ingestion + search (used for precision@5 benchmark).
• Developer onboarding: docs/dev-setup.md with steps to run locally (Linux/macOS/WSL), run tests, and extend plugins.
• Implementation backlog: child work items or issue templates for each MVP story, migration notes, and sample CLI usage examples.

Related work

• README.md: project overview and references. (README.md)
• Review OB1 Life Engine recipe and recommend OpenBrain actions (OB-0MN8OAZVM00890DW)
• Improve agents and skills to send completion summaries to OpenBrain (OB-0MN8O6ID4009WHAQ)
• Define and implement Brainstorming (OB-0MN8N4FCO002UN3D)
• Define and implement Memory Compartmentalization (OB-0MN8O3NC0004RPH6)

Appendix: Clarifying questions & answers

• Q: "What is the primary goal for this PRD? Pick one (recommended choice first)." — Answer (user): "Product vision + longer roadmap". Source: interactive reply. Final: yes.

• Q: "Who is the primary audience for the PRD? (select one)" — Answer (user): "Developers & engineers (Recommended)". Source: interactive reply. Final: yes.

• Q: "What timeframe/scope should the PRD cover? (select one)" — Answer (user): "Mid — 6 months". Source: interactive reply. Final: yes.

• Q: "Which deployment/architecture should the PRD assume for the 6‑month roadmap? Pick one." — Answer (user): "Local‑first CLI only". Source: interactive reply. Final: yes.

• Q: "Which three items should be the highest priority to include in the PRD's 6‑month plan? Pick up to 3." — Answer (user): "Local ingestion & embedding pipeline, Fast semantic search & retrieval, Scheduled agents / brainstorming workflows". Source: interactive reply. Final: yes.

• Q: "Which 2–3 success metrics should the PRD target for the 6‑month horizon? Pick items and add numeric targets if possible." — Answer (user): "Search relevance / quality, Feature delivery milestones (Recommended)". Source: interactive reply. Final: yes.

• Q: "I drafted three candidate user stories for the PRD — please pick one: accept as-is, accept with edits (you'll paste edits), or provide your own." — Answer (user): "Accept as-is (Recommended)". Source: interactive reply. Final: yes.

• Q: "Choose numeric targets for the two selected success metrics (search relevance, feature delivery) or provide custom targets." — Answer (user): "Preset B — ambitious" (precision@5 ≥ 0.8; deliver 10 stories with ≥90% tests passing). Source: interactive reply. Final: yes.

• Q: "Confirm technical defaults for the 6‑month PRD (pick one). These are conservative, local‑first choices consistent with README." — Answer (user) initial: "Accept recommended defaults (Recommended)" (SQLite + sqlite-vec, CPU-friendly models, Linux/macOS). Source: interactive reply. Revision: user later relaxed the earlier 'no network' constraint; see next entry. Final: updated.

• Q: "Relax the no network access constraint. Many features will use the nextwork. What we do not want to do is rely on any proprietary service on the network for core functionality." — Answer (user): "Allow network access for features, but do not rely on proprietary services for core functionality; ensure any networked provider/hosted model is optional, pluggable, and has documented local fallback." Source: interactive reply. Prior: earlier draft said 'No network access required by default'. Final: yes.

Open Questions

• OPEN QUESTION: Do we want Windows support in the 6‑month MVP? (asked to user; awaiting answer)

Draft created by Map on behalf of requester.

[SorraTheOrc]

Map

Related work report appended by find_related: README, OB1 review, Brainstorming, Memory Compartmentalization, agents/skills improvements (see PRD description).

[SorraTheOrc]

Map

Effort & risk estimate (preliminary) attached: T-shirt XL, recommended ~360h, primary risks: security, CLI integration, test coverage. See skill output.

[SorraTheOrc]

Map

User approved the intake draft. Proceeding to five review passes and finalization.

[SorraTheOrc]

Map

Completed five intake reviews: Completeness, Capture fidelity, Related-work & traceability, Risks & assumptions, Polish & handoff. No intent-changing edits made; conservative clarifications added.

[SorraTheOrc]

OpenCode

Created docs/PRD.md, docs/dev-setup.md, acceptance test scaffold and sample dataset. Commit a97daf3.

[SorraTheOrc]

OpenCode

Automated dependency chain added between epic children:

• OB-0MN9HP45R003RQ19 -> OB-0MN9HP0EC0065TYT: OK
• OB-0MN9HP5Y30049NKB -> OB-0MN9HP45R003RQ19: OK
• OB-0MN9HP7TO003G7IB -> OB-0MN9HP5Y30049NKB: OK
• OB-0MN9HVEGW006EDBG -> OB-0MN9HP7TO003G7IB: OK
• OB-0MN9HWGAL001452N -> OB-0MN9HVEGW006EDBG: OK
• OB-0MN9HWGOE002YR1I -> OB-0MN9HWGAL001452N: OK
• OB-0MN9HWH4D009EST6 -> OB-0MN9HWGOE002YR1I: OK
• OB-0MN9HWHHE0081ABP -> OB-0MN9HWH4D009EST6: OK
• OB-0MN9HWHUK009CH83 -> OB-0MN9HWHHE0081ABP: OK
• OB-0MN9HWI8X007ITRF -> OB-0MN9HWHUK009CH83: OK
• OB-0MN9HWIMJ0080RUU -> OB-0MN9HWI8X007ITRF: OK
• OB-0MN9HWJ0C009QZI2 -> OB-0MN9HWIMJ0080RUU: OK
• OB-0MN9HWJFP0000VWI -> OB-0MN9HWJ0C009QZI2: OK
• OB-0MN9HWJU90049W1V -> OB-0MN9HWJFP0000VWI: OK
• OB-0MN9IK91K003S6IM -> OB-0MN9HWJU90049W1V: OK
• OB-0MN9IKBB1002VK3U -> OB-0MN9IK91K003S6IM: OK

[SorraTheOrc]

OpenCode

Plan consolidated: Reduced from 20+ work items to 10 core MVP items. Closed 11 duplicates: provisioning tasks merged into SQLite, embedding children merged into adapter, ingest children merged into CLI, tests merged into benchmark, docs merged into onboarding, templates merged into plugin task. Created separate epic OB-0MND1YHKL00416L9 for import integrations (chat importers). Updated dependency chain to reflect DB-first flow. See consolidated-plan.md for full details.

[SorraTheOrc]

OpenCode

Fixed circular dependency: Removed bidirectional dependency between Embedding adapter (OB-0MN9HWGOE002YR1I) and Ingest CLI (OB-0MN9HWGAL001452N). The correct flow is: SQLite (foundation) → Embedding adapter → Ingest CLI (higher-level). Also removed SQLite dependency on Embedding adapter (was backwards) and removed Ingest CLI dependency on completed task OB-0MN9HVEGW006EDBG. Dependency chain is now: SQLite → Embedding → Ingest → Search/Benchmark/Scheduler.