Product Forge

Product Intelligence, Analytics Intelligence & Agentic Workspace — a specialist Forge in the Forge ecosystem for Product Managers, Product Analysts and technical teams.

What it is

Product Forge turns scattered product knowledge (docs, metrics, APIs, data sources, decisions) into a governed, queryable, agent-ready context:

• Product Context Packages — curated per-product context: entities, sources, agents, data contracts.
• Knowledge & Graph — ingestion → chunking → FTS retrieval with citations → entity/relation graph.
• Agent Factory — define, evaluate, publish product-specialist agents with scoped permissions.
• Analytics Studio — read-only-by-default governed SQL over registered data sources (NL, visual builder, advanced editor sharing one contract).
• Dashboard Studio — real-data KPI cards, charts, filters, sharing.
• Planning & Discovery — opportunities, hypotheses, prioritization, PRDs, roadmaps, decisions with evidence traceability.
• Product Operations — scheduled reports, watches, alerts, approvals.

Surfaces: CLI (productforge), HTTP API (/api/v1, forge/PortalApi/1 envelope), MCP server (productforge mcp serve), Forge Portal (productforge portal).

Install

Bootstrap the CLI (wheel → isolated venv → launcher):

./install.sh          # POSIX · ./install.windows.sh · ./install.ps1
# or dev flow: uv venv .venv && uv pip install -p .venv/bin/python -e ".[dev]"

Then install the Forge. For the non-technical first-run path, use one command:

productforge install

It applies the recommended local profile, installs the Product Forge PM/PA portal assets and host integrations, records a receipt and runs diagnostics. It also provisions a managed PostgreSQL container for the recommended profile. It does not open a browser or ask for AI provider credentials.

The recommended profile uses an installation-scoped Docker volume and a generated local DSN. productforge portal reconciles the database before opening the UI; productforge status and productforge doctor report its health. Use --profile minimal for an explicit SQLite-only offline install.

On a TTY, advanced users can use the guided wizard (same contract as sparkforge-aws install):

1. Scope — project | workspace | user
2. Install profile — minimal | recommended | full
3. Execution profile — economy | balanced | deep
4. Components — skills, agents, mcp, tui, graph-studio
5. Hosts — claude, devin, codex, copilot | all | none
6. Real dry-run plan → confirm → apply → doctor

Flags keep CI non-interactive JSON:

productforge install --yes --scope project --host claude,devin \
  --components skills,agents,mcp --gateway-profile balanced
productforge install --dry-run                    # plan only, zero writes
productforge repair / update / uninstall          # lifecycle verbs (ledger-governed)
productforge status

Install is governed by the vendored forge installkit (src/product_forge/_installkit.py, canonical the-forge/scripts/installkit) — the same engine spark-forge-aws uses: profile-gated assets, ledger-tracked ownership (uninstall removes only owned files), AGENTS.md marker blocks, .mcp.json registration, drift detection and repair. The 16 pf-* skills ship inside the wheel (force-include) into each selected host's skills dir; content bundles (rules, evals, knowledge) land under <state_dir>/bundle/.

Full guide: docs/INSTALL.md.

Quickstart

productforge workspace create "My Company"
productforge product create <workspace_id> "Payments" --domain fintech
productforge knowledge import <product_id> docs/payments.md
productforge knowledge query <product_id> "retry policy"
productforge agent create <product_id> "Payments Analyst" --template product-analyst
productforge api serve            # local API on 127.0.0.1
productforge portal               # opens the Product Forge PM/PA portal and manages the local Agent Bridge
productforge mcp serve            # MCP tools for agent hosts

Ecosystem compatibility

• forge/SpecialistAgenticManifest/v1 → forge.agentic.json (Forge Protocol provider).
• forge/PortalApi/1 envelope on every API response (schema + correlation_id + data|error).
• Vendored _forge_portal.py (byte-parity with forge-portal/portal_core/forge_portal.py) + ProductForgePortal adapter.
• Read-only by default; mutations need the local bearer token (productforge api token --show).

Skills

skills/ ships 16 domain skills (§15.3: strategy, discovery, prioritization, requirements, SQL, analytics, dashboards, AWS, Azure, APIs, data products, governance, FinOps, reporting, knowledge, agent factory) — each validated with skill-creator (agents/openai.yaml + quick_validate clean). Index: skills/index.json.

MCP

productforge mcp serve — stdio JSON-RPC server exposing 14 tools: productforge_status, product_list, product_context, knowledge_query, agent_list, product_graph, knowledge_health, query_validate, dashboard_list, roadmap, alerts_list, audit_list, agent_run, agent_eval. install --scope project registers it in .mcp.json for host auto-discovery.

Status

Beta — Cycles 1–18 complete: foundation → knowledge/graph → agent runtime → analytics → dashboards → planning → cloud/API/data-product intelligence → operations → advanced agentic → homologation → Portal studio → real LLM runtime → semantic metrics → health/evidence/experiments → persona homologation. The Payments Product Intelligence scenario runs end-to-end (tests/test_homologation.py) plus a negative-scenario suite. See docs/audit/ for the ecosystem audit, ADRs and the readiness report; deferred items are listed there explicitly.

Docs