- Decision report: status flipped to MIGRATED; added migration log with
five WSL2 gotchas (bzip2 missing, no official Photon image,
firewall=true blocks cross-LAN, vmIdleTimeout=-1 ineffective,
PowerShell pre-expansion of bash $(...)) and resource snapshot.
- mana-geocoding CLAUDE.md: PHOTON_SELF_API_URL note now reflects live
primary status on mana-gpu since 2026-04-28.
- photon-self/: operator scripts for the weekly DB refresh — update.sh
(atomic-swap with rollback), systemd unit + timer (Sun 03:30 +30min
jitter, Persistent=true), README with re-installation instructions
for DR. Currently installed and enabled on mana-gpu.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Compares Pelias / Nominatim / Photon for self-hosting on the GPU
server, with current (2026-04-28) numbers from upstream docs +
GraphHopper's Photon-data downloads:
Photon Europe pre-built dump: 30.6 GB, weekly refresh
Photon Germany pre-built dump: 5.8 GB, weekly refresh
Nominatim Germany import: ~100 GB disk, 8–12 h, 12 GB RAM
Pelias DACH (current): 3 GB RAM, 4 services, JS patch hack
Recommendation: Photon Europe-wide on mana-gpu. Single Java process,
embedded OpenSearch, no PBF import (download a tarball, restart),
weekly auto-updates from GraphHopper, integrates with the wrapper's
existing PhotonProvider via just an env-var change.
Once self-hosted, Photon registers as `privacy: 'local'` — the
sensitive-query block (Hausarzt, Klinikum, …) gets a real local
backend and no longer has to return empty results when Pelias is
down. Public Photon stays in the chain as a `privacy: 'public'`
last-resort fallback.
Migration plan included (~3–4 h total, ~1 h waiting), with
phase-by-phase risk assessment.
Pelias does not return — the 3 GB RAM + multi-container + patched
JS combination has no operational case once we have a self-hosted
Photon that already matches our wrapper's wire format.
Three Claude-Code-inspired primitives for runPlannerLoop, derived from the
reverse-engineering reports in docs/reports/:
1. **Policy gate** (@mana/tool-registry) — evaluatePolicy() gates every tool
dispatch: denies admin-scope, denies destructive tools not in the user's
opt-in list, rate-limits per tool (30/60s default), flags prompt-injection
markers in freetext without blocking. Wired into mana-mcp with a
per-user rolling invocation log and POLICY_MODE env (off|log-only|enforce,
default log-only). mana-ai uses detectInjectionMarker only — tool dispatch
there is plan-only, so rate-limit/destructive checks don't apply yet.
2. **Reminder channel** (packages/shared-ai/src/planner/loop.ts) — new
reminderChannel callback in PlannerLoopInput. Called once per round with
LoopState snapshot (round, toolCallCount, usage, lastCall); returned
strings wrap in <reminder> tags and inject as transient system messages
into THIS LLM request only. Never pushed to messages[] — the Claude-Code
<system-reminder> pattern that keeps the KV-cache prefix stable.
3. **Parallel reads** (loop.ts) — isParallelSafe predicate enables
Promise.all dispatch when every tool_call in a round is parallel-safe,
in batches of PARALLEL_TOOL_BATCH_SIZE=10. Any non-safe call downgrades
the whole round to sequential. messages[] always appends in source
order, never completion order, so the debug log stays linear.
Default-off (undefined predicate) preserves pre-M1 behaviour.
Tests: 21 new in tool-registry (policy), 9 new in shared-ai (5 parallel,
4 reminder). All 74 green, type-check clean across 4 packages.
Design/plan: docs/plans/agent-loop-improvements-m1.md
Reports: docs/reports/claude-code-architecture.md,
docs/reports/mana-agent-improvements-from-claude-code.md
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Capture the surprises from the first deploy so the next rollout
(or rollback) has the full picture without spelunking logs:
- mana-research had never been started on the Mac-Mini, even though
it was defined in compose. First-boot via `docker compose up -d`.
- research.* schema is not auto-migrated on service boot — drizzle
push must be triggered explicitly: `docker exec mana-research
bun run db:push`. 5 tables created.
- GOOGLE_GENAI_API_KEY was missing in /Users/mana/.../mana-monorepo/.env.
Copied the local key over, with `.env.bak.pre-gemini-deep-research`
as rollback anchor.
- Redis NOAUTH fix (commit 4867300d0) referenced here.
- Smoke-test outcome documented: the 500 was mana-credits 404 on a
test user without a wallet row — expected, and it proves the whole
auth/dispatch chain up to the credits hop works.
- Also noted: mana-llm has the same bare REDIS_URL in compose
(out-of-scope for this deploy), and /providers/health does not list
async providers (known design gap).
Status header updated to reflect deploy completion. Flag stays off
(MANA_AI_DEEP_RESEARCH_ENABLED=false) pending explicit enablement.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Opt-in path for missions that want Gemini Deep Research Max (up to 60 min
per task) instead of the shallow RSS pre-research. Because Max runs well
past a single 60-second tick, the state is carried across ticks:
tick N: submit → INSERT mission_research_jobs row → skip planner
tick N+k: poll → still running → skip planner (metric pending_skips)
tick N+m: poll → completed → inject as ResolvedInput, DELETE row, plan
- ManaResearchClient talks to mana-research's new internal
/v1/internal/research/async endpoints with X-Service-Key +
X-User-Id. Graceful-null on transport errors so a flaky
mana-research never crashes the tick loop.
- New table mana_ai.mission_research_jobs with PK (user_id, mission_id)
— presence is the "pending" flag; delete-on-terminal keeps queries
trivial.
- handleDeepResearch() encapsulates the state machine; planOneMission
now returns a discriminated union (planned | skipped | failed) so
"research pending" isn't miscounted as a parse failure.
- Opt-in at TWO gates to keep cost in check ($3–7/task, 1500 credits
per run):
1. MANA_AI_DEEP_RESEARCH_ENABLED=true server-side (default off)
2. DEEP_RESEARCH_TRIGGER regex matches the mission objective
(strict: "deep research", "tiefe recherche", "umfassende
recherche", "hintergrundrecherche", "deep dive")
Falls back to shallow RSS when either gate fails or the submit
errors upstream.
- Prom metrics: mana_ai_research_jobs_{submitted,completed,failed}_total
labelled by provider, plus _pending_skips_total.
- docker-compose wires MANA_RESEARCH_URL + the opt-in flag and adds
mana-research to depends_on.
- Full write-up with real API response shape (outputs plural, not
OpenAI-style), step-3 MCP-server plan (security-gated, not built),
ops + kill-switch: docs/reports/gemini-deep-research.md.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Competitive analysis of ClubDesk (reeweb ag, ~20'000 DACH clubs) with a
dual-use roadmap identifying features that benefit both clubs and general
users (freelancers/creators). First chosen step: invoices module with
Swiss QR-Bill as the CH-differentiator.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
New Bun/Hono service on port 3068 that bundles many web-research providers
behind a unified interface for side-by-side comparison. All eval runs
persist in research.* (mana_platform) so quality can be reviewed later.
Providers (Phase 1+2):
search: searxng, duckduckgo, brave, tavily, exa, serper
extract: readability (via mana-search), jina-reader, firecrawl
Endpoints:
POST /v1/search, /v1/search/compare — single + fan-out
POST /v1/extract, /v1/extract/compare — single + fan-out
GET /v1/runs, /v1/runs/:id — history
POST /v1/runs/:run/results/:id/rate — manual eval
GET /v1/providers, /v1/providers/health — catalog + readiness
Auto-routing: when `provider` is omitted, queries are classified via regex
(fast path, 0ms) with optional mana-llm fallback, then routed to the first
available provider for that query type (news → tavily, academic → exa,
semantic → exa, etc.).
Credits: server-key calls go through mana-credits reserve → commit/refund
so failed provider calls don't charge the user. BYO-keys supported via
research.provider_configs (UI arrives in Phase 4).
Cache: Redis with graceful degradation (1h TTL for search, 24h for
extract). Pay-per-use APIs only — no subscription-gated providers.
Docs: docs/plans/mana-research-service.md + docs/reports/web-research-capabilities.md
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Fix missing strikethroughs in §6 table (#1, #2, #6) and update Fazit
to reflect final state: 7 of 10 items done. Document remaining 3
langfristige Punkte with context on dependencies and priorities.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Complete tool handler coverage for the MCP server:
Todo: complete_tasks_by_title
Calendar: create_event (with timeBlock)
Notes: update_note, append_to_note, add_tag_to_note
Places: create_place, visit_place, get_places
Drink: log_drink, get_drink_progress, undo_drink
Food: log_meal, nutrition_summary
Journal: create_journal_entry
Habits: create_habit, log_habit (get_habits improved)
News: save_news_article
27 of 29 tools now have real implementations. Remaining 2
(research_news, get_current_location) need external service
calls that aren't available in the API server context.
Also updates architecture comparison report to mark MCP as done.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Enable real-time token streaming during the planner "calling-llm" phase
so the user sees live progress ("empfange Plan… 128 tokens") instead of
a static spinner. The parser still receives the full text once complete —
no partial-JSON risk.
Changes:
- Extract shared SSE parser from playground into @mana/shared-llm/sse-parser
- remote.ts: use stream:true when onToken callback is provided
- AiPlanInput: add optional onToken field (shared-ai)
- ai-plan task: pass onToken through to backend.generate()
- runner.ts: throttled (500ms) phaseDetail updates during streaming
- Playground: refactored to use shared SSE parser
Also includes: AI agent architecture comparison report (docs/reports/)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
