Skip to content

Runtime Roadmap — Lean g-agent Runtime

Last updated: 2026-02-12

This roadmap tracks runtime priorities for g-agent, while keeping it lightweight, cross-platform, and operator-controlled.

North Star

Build a personal assistant that is:

  • proactive (can initiate useful work),
  • memory-strong (recalls user context reliably across sessions),
  • workflow-capable (Google + browser + channel orchestration),
  • safe by default (strict boundaries for personal and guest modes).

Scope Boundaries

This roadmap is focused-scope.

In scope:

  • high-value runtime behavior
  • memory quality
  • proactive orchestration
  • security posture for mixed personal/guest usage
  • observability for daily operations

Out of scope:

  • heavy framework expansion
  • channel sprawl without clear reliability/ops value
  • large control-plane abstractions that reduce local clarity

Delta Tracks and Status

1) Execution Runtime with Checkpoints

Status: implemented (v1)

Implemented:

  • task lifecycle checkpoints (plan -> execute -> verify -> reflect -> done)
  • persisted task state in workspace (state/tasks/*.json)
  • restart-safe completion tracking

Code:

  • backend/agent/g_agent/agent/runtime.py
  • backend/agent/g_agent/agent/loop.py

2) Memory Retrieval Intelligence

Status: implemented (v1)

Implemented:

  • memory metadata schema (type, confidence, source, last_seen, supersedes)
  • dedup/supersede behavior
  • ranked recall via confidence + recency + relevance

Code:

  • backend/agent/g_agent/agent/memory.py

3) Proactive Agent Engine

Status: implemented (v1)

Implemented:

  • perfect-day schedule prompts
  • calendar watch lead-time reminders
  • quiet-hours gating + dedupe state

Code:

  • backend/agent/g_agent/proactive/engine.py
  • backend/agent/g_agent/cli/commands.py
  • backend/agent/g_agent/cron/service.py

4) Multimodal Output

Status: implemented (v1)

Implemented:

  • outbound message tool supports text/image/voice/sticker/document
  • generated voice + sticker fallback from plain text
  • workflow-pack media flags (--voice, --image, --sticker)

Code:

  • backend/agent/g_agent/agent/tools/message.py
  • backend/agent/g_agent/agent/workflow_packs.py

5) Workflow Packs (Google-first)

Status: implemented (v1)

Implemented packs:

  • daily_brief
  • meeting_prep
  • inbox_zero_batch

Implemented scope:

  • intent-level orchestration prompting
  • multimodal output options
  • media-first --silent mode

Code:

  • backend/agent/g_agent/agent/workflow_packs.py
  • backend/agent/g_agent/agent/loop.py

6) Guest Safety Boundary

Status: implemented (v1)

Implemented:

  • policy presets: personal_full, guest_limited, guest_readonly
  • channel/sender scoped policy map
  • deny-by-default behavior for guest restrictions

Code:

  • backend/agent/g_agent/config/presets.py
  • backend/agent/g_agent/agent/loop.py
  • backend/agent/g_agent/cli/commands.py

7) Observability and Evaluation Harness

Status: implemented (v1)

Implemented:

  • local metrics event sink
  • tool/cron reliability snapshots
  • latency + success-rate surfaces in status/doctor paths

Code:

  • backend/agent/g_agent/observability/metrics.py
  • backend/agent/g_agent/cli/commands.py

What Remains (Hardening Backlog)

Recently Completed

  • Telegram/WhatsApp reconnect harness coverage:
  • backend/agent/tests/test_channel_reconnect.py
  • OAuth edge-case regression checks added for expired refresh token and scope drift:
  • backend/agent/tests/test_google_oauth_edges.py
  • backend/agent/g_agent/agent/tools/google_workspace.py
  • Provider-specific retry taxonomy for transient tool failures:
  • backend/agent/g_agent/agent/loop.py
  • backend/agent/tests/test_retry_and_idempotency.py
  • Memory quality fixtures and ranking consistency assertions:
  • backend/agent/tests/fixtures/memory_conflicts.md
  • backend/agent/tests/test_memory_intelligence.py
  • backend/agent/g_agent/agent/memory.py
  • Multilingual overlap fixture coverage + summary/fact drift checks:
  • backend/agent/tests/fixtures/memory_multilingual.md
  • backend/agent/tests/test_memory_intelligence.py
  • backend/agent/g_agent/agent/memory.py
  • Metrics export path + dashboard-friendly scrape summary:
  • backend/agent/g_agent/observability/metrics.py
  • backend/agent/g_agent/cli/commands.py
  • backend/agent/tests/test_observability_metrics.py
  • Optional lightweight HTTP /metrics endpoint mode (disabled by default):
  • backend/agent/g_agent/observability/http_server.py
  • backend/agent/g_agent/cli/commands.py
  • backend/agent/tests/test_metrics_http_server.py
  • Metrics retention/pruning controls for events.jsonl growth management:
  • backend/agent/g_agent/observability/metrics.py
  • backend/agent/g_agent/cli/commands.py
  • backend/agent/tests/test_metrics_retention_alerts.py
  • Alert-threshold summary output for uptime/SLO monitoring:
  • backend/agent/g_agent/observability/metrics.py
  • backend/agent/g_agent/cli/commands.py
  • backend/agent/tests/test_metrics_retention_alerts.py
  • Integration-level reconnect harness (manager + dispatcher + multi-channel recovery):
  • backend/agent/tests/test_channel_reconnect.py
  • backend/agent/g_agent/channels/manager.py
  • Semantic normalization checks for mixed-language recall drift:
  • backend/agent/g_agent/agent/memory.py
  • backend/agent/tests/test_memory_intelligence.py
  • Cross-scope memory conflict regression checks (profile/long-term/custom):
  • backend/agent/g_agent/agent/memory.py
  • backend/agent/tests/test_memory_intelligence.py
  • Operator-facing memory audit diagnostics (memory-audit + doctor checks):
  • backend/agent/g_agent/cli/commands.py
  • backend/agent/g_agent/agent/memory.py
  • backend/agent/tests/test_memory_intelligence.py
  • Operator-facing security baseline audit automation (security-audit + doctor summary):
  • backend/agent/g_agent/security/audit.py
  • backend/agent/g_agent/cli/commands.py
  • backend/agent/tests/test_security_audit.py
  • Operator-facing security baseline auto-remediation helper (security-fix dry-run/apply):
  • backend/agent/g_agent/security/fix.py
  • backend/agent/g_agent/cli/commands.py
  • backend/agent/tests/test_security_fix.py
  • Channel manager long-running supervisor restarts for unexpected channel crashes:
  • backend/agent/g_agent/channels/manager.py
  • backend/agent/tests/test_channel_reconnect.py
  • Outbound dispatch retries with capped exponential backoff for transient channel-send failures:
  • backend/agent/g_agent/channels/manager.py
  • backend/agent/tests/test_retry_and_idempotency.py
  • Metrics alert compact summaries surfaced in status/doctor plus Prometheus alert gauges:
  • backend/agent/g_agent/observability/metrics.py
  • backend/agent/g_agent/cli/commands.py
  • backend/agent/tests/test_metrics_retention_alerts.py
  • Scoped tool-policy identity normalization + guardrail audit checks:
  • backend/agent/g_agent/agent/loop.py
  • backend/agent/g_agent/security/audit.py
  • backend/agent/tests/test_policy_presets.py
  • backend/agent/tests/test_security_audit.py
  • Channel supervisor restart-burst cooldown guard for long crash loops:
  • backend/agent/g_agent/channels/manager.py
  • backend/agent/tests/test_channel_reconnect.py

P0 — Reliability Gaps (next)

  1. none (closed)

P1 — Memory Quality

  1. none (closed)

P2 — Observability Ops

  1. none (closed)

Exit Criteria for Delta Phase

Delta phase is considered complete when:

  • all P0 items are implemented and covered by automated tests
  • memory regression fixtures are in CI and stable
  • one optional metrics export path is available and documented
  • production checklist in backend/agent/SECURITY.md remains valid without exceptions

Relationship to Main Docs

  • product-level narrative: README.md
  • backend setup and operations: backend/agent/README.md
  • security posture and hardening: backend/agent/SECURITY.md

This roadmap only tracks the remaining g-agent runtime work.