Skip to content

ROADMAP.md

Latest commit

 

History

History
200 lines (149 loc) · 8.54 KB

ROADMAP.md

File metadata and controls

200 lines (149 loc) · 8.54 KB
Raw

Roadmap

Current Product Direction

Build parallel-agents into the execution engine for Parallel Agents Office — a no-code AI software company OS that takes a project from idea to release through a coordinated, legible, and (ultimately) forkable org of AI agents.

The near-term product path is:

Stabilize engine -> First-class company workflows -> GitHub approval workflow
  -> Local gateway/job system -> Local desktop office -> Hosted MCP

Phase 0: Engine Stabilization

Status: mostly complete.

Completed:

  • CLI, MCP server, PyPI package, npm wrapper, and core worker engine.
  • Evidence storage with file and SQLite backends.
  • Product vision, PR/FAQ, operating model, quality bar, tech stack policy, and workflow docs.
  • Release hygiene docs and compatibility notes.
  • Consolidated release checklist command (parallel-agents release verify).

Remaining:

  • Keep package metadata, changelog, README, and examples synchronized for each release.

Phase 1: Company Workflow Models

Status: mostly complete.

Completed:

  • ProductBrief
  • PRFAQDocument
  • TechStackDecision
  • ArchitectureRFC
  • RoadmapPlan
  • SprintPlan
  • ReleaseReadinessReport
  • PostReleaseReview
  • CLI command group: parallel-agents company

Remaining:

  • Expand generated artifacts with richer user context and research inputs.
  • Add more workflow-specific validation as real users exercise the commands.

Phase 2: GitHub-First No-Code Workflow

Status: partially complete.

Completed:

  • Create issue plans from roadmap artifacts.
  • Approval-gated company plan -> approve -> apply flow.
  • Immutable approval audit log entries.
  • Apply-time policy checks for repositories, labels, and milestones.
  • GitHub labels/milestones/branch templates.
  • Stable branch name generation.
  • PR summary generation from stored run outputs.
  • Draft PR creation from run-linked summaries (company pr-create).
  • PR summary/risk GitHub comment posting (company pr-comment).
  • Run-to-PR linking artifacts and immutable PR-link audit events.
  • GitHub label/milestone synchronization commands (company sync-labels, company sync-milestones).

Remaining:

  • Deeper GitHub workflow automation beyond issue/PR surface.

Phase 3: Gateway and Job System

Status: mostly complete.

Goal: move from one-shot CLI runs to persistent local project sessions.

Completed:

  • Optional local gateway dependency group.
  • Local FastAPI gateway command.
  • SQLite project/run/job/event store.
  • Basic company idea, roadmap, plan, approve, apply, artifacts, and events endpoints.
  • In-process job queue with persistent run/job status tracking.
  • Cancellation and retry controls for queued/failed/policy-blocked runs.
  • Run listing and per-run job inspection endpoints.
  • Incremental run event stream for enqueue/start/status/complete/cancel.
  • Optional API-key protection for non-local gateway exposure.
  • Optional JWT HS256 auth mode with issuer/audience validation for non-local gateway exposure.
  • Gateway-managed core pipeline runs via POST /runs/pipeline, with persistent run/job/event state and final-output artifacts.

Remaining:

  • Deeper desktop/exe shell integration on top of the gateway control plane.
  • Hosted-grade auth for multi-tenant remote deployments (OAuth/session model).

Phase 4: Local Desktop Office

Status: in progress.

Goal: local .exe experience that works inside the selected project folder.

Completed:

  • Project-folder workspace layout under .parallel-agents/.
  • parallel-agents office init to create local workspace metadata and directories.
  • parallel-agents office onboard to create/repair workspace setup and report model/GitHub readiness for first use.
  • parallel-agents office status to inspect workspace health.
  • parallel-agents office doctor to run local readiness diagnostics and strict checks.
  • parallel-agents office fix-setup for safe CLI remediation fallback.
  • Desktop project-home summary and recent project picker.
  • Desktop project-home card now surfaces doctor health status and warning/failure counts.
  • Desktop Projects page now provides one-click Onboard, Run Doctor, and Fix Setup actions.
  • Desktop approvals queue filters with approved issue-plan apply action.
  • Desktop GitHub PR creation with run-linked PR summary artifact.
  • Workspace knowledge layer v1:
    • office memory add/list/search/policies
    • persisted memory files under .parallel-agents/memory/
    • gateway memory APIs for entry/policy access
  • Desktop Runs page now executes real pipeline runs with live activity streaming and worker status updates.
  • Desktop project-home release/productivity card from eval artifacts (impact, acceptance, regression, gate, cost, duration).
  • Desktop metric-history timeline with latest-vs-previous deltas from evaluation score artifacts.
  • Desktop filtered trend view (overall/project/workflow slices, date windows, metric selection, inline trend rendering).
  • Desktop exportable trend reporting (CSV + Markdown from current filtered view).
  • Desktop graphical trend chart rendering with PNG export.
  • Desktop cross-run benchmark comparison panel (baseline vs candidate) with delta report export.
  • Desktop drill-down analytics in comparison view (workflow/project/case-level change drivers).
  • Workflow navigation into detailed case evidence and artifact links from comparison rows.
  • Desktop run execution can route through the local gateway control plane when available, with in-process fallback.
  • Desktop can start, stop, and inspect a project-scoped local gateway process.
  • Desktop worker tiles now consume structured pipeline trace events, with legacy status-string fallback.
  • Local gateway channel-adapter boundary with default pairing/allowlist behavior for inbound senders.
  • Signed Slack Events endpoint mapped onto the local pairing adapter.
  • CLI operator commands for channel inbound simulation, pairing approval, and peer listing.
  • Generated patches require final-output approval before desktop PR creation.
  • Desktop PR branch suggestions avoid pushing directly from main/master.
  • PyInstaller spec includes project-office module for standalone binary builds.

Remaining:

  • Desktop polish and usability hardening for production .exe workflows.
  • Installable OS-level gateway daemon/tray lifecycle beyond the current desktop-owned session process.
  • Telegram connector on top of the pairing adapter.
  • Discord connector after Telegram, once channel/server routing rules are clearer.

Phase 5: Remote MCP Product Surface

Status: partially complete.

Goal: use Parallel Agents Office inside Claude, ChatGPT, Codex, Cursor, and similar tools.

Completed:

  • MCP tool discovery surface (tool_discovery) with read/write classification and approval-gate metadata.
  • Gateway-hosted MCP tool surface:
    • GET /mcp/tools for capability discovery
    • POST /mcp/tools/{tool_name} for remote tool invocation
    • remote write-tool policy gate (PA_GATEWAY_ALLOW_REMOTE_WRITE_TOOLS / --allow-remote-write-tools)
  • Workspace access audit trail for gateway traffic (GET /audit/access).

Remaining:

  • Hosted MCP endpoint.
  • OAuth.
  • Remote public read-only access model with tenant/session identities.
  • Remote write tools behind approval with OAuth/session claims.

Phase 6: Proof and Evaluation

Status: partially complete.

Completed:

  • Evaluation dataset format.
  • parallel-agents eval run.
  • parallel-agents eval score.
  • parallel-agents eval compare for baseline vs candidate delta reporting.
  • parallel-agents eval gate for CI/release threshold enforcement.
  • parallel-agents eval annotate for run-level acceptance/regression/finding updates.
  • parallel-agents eval sync-pr for GitHub PR acceptance ingestion.
  • parallel-agents eval sync-ci for CI outcome regression ingestion.
  • parallel-agents eval breakdown for cost/time views by project and workflow.
  • Scorecards for speed gain, acceptance rate, regression rate, finding precision, and weighted delivery impact.
  • Starter public benchmark dataset (examples/public_benchmark_v1.json).
  • Expanded public benchmark dataset (examples/public_benchmark_v2.json).

Remaining:

  • Continue broadening benchmark coverage and publish external benchmark baselines.
  • PR acceptance tracking is integrated through eval sync-pr and PR link files.
  • Regression tracking from CI outcomes is integrated through eval sync-ci.
  • Cost/time views by workflow and project are integrated through eval breakdown.

Near-Term Priority

The next high-impact work is to expand the local desktop office into a practical .exe product surface:

  • project picker and local workspace home
  • artifact browser and approval queue
  • GitHub connection and PR flow integration