jo-inc Open Source Tools → OiMy Review

Date: 2026-05-29

Scope

Reviewed the askjo.ai build narrative available via public site/GitHub and jo-inc open source repositories:

  • https://github.com/jo-inc/camofox-browser
  • https://github.com/jo-inc/pi-mem
  • https://github.com/jo-inc/pi-reflect
  • https://github.com/jo-inc/safe-skill-search
  • https://github.com/jo-inc/wa_meow
  • https://github.com/jo-inc/skills
  • https://github.com/jo-inc/skills-db

    • Note: x.com direct fetch of the referenced post failed due x.com page restrictions, so this review uses the public repo READMEs + askjo.ai site copy. The key claimed themes still align: local/private data, gbrain-like memory, multi-channel access, dedicated user cloud machine, Mac + Telegram/WhatsApp/email.

    ---

    Executive recommendation

    There are three genuinely useful patterns/tools for OiMy:

    1. camofox-browser — useful for authorized browser automation where Playwright/Chrome gets blocked or needs stable accessibility snapshots. Adopt as an optional browser backend, not as a default evasion layer.

    2. pi-reflect pattern — very useful: a scheduled reflection loop that reads transcripts/evals and edits behavioral files/skills to prevent repeated failures. We should adapt this into OiMy’s skill QA and agent-maintenance pipeline.

    3. safe-skill-search pattern — useful for a better skills marketplace/library search with quality scoring. OiMy already has a skills library; this gives us a path to quality-ranked search, curation, and skill discovery.

    Useful but not immediate:

  • **wa_meow** could be a future self-hosted WhatsApp bridge, but OiMy already has Twilio WhatsApp infrastructure. Evaluate later because linked-device WhatsApp bridges carry reliability and ToS/session risks.
  • **pi-mem** overlaps heavily with OpenClaw/GBrain and OiMy memory. Use its simple file-memory UX patterns, not the package itself.

    • Not a priority:

  • jo-inc skills / skills-db are useful references, not direct infrastructure wins.

    • ---

    Tool-by-tool review

    1. camofox-browser

    What it is: Anti-detection browser server for AI agents, wrapping Camoufox with REST APIs, accessibility snapshots, stable element refs, session isolation, cookie import, VNC login, search macros, YouTube transcripts, structured extraction, screenshots, traces.

    Why it matters for OiMy:

    OiMy has ambitious “do it for me” workflows: forms, bookings, shopping, school portals, trip planning, medical billing, etc. Regular headless Chrome/Playwright often gets blocked or breaks on consumer websites. camofox-browser gives an agent-oriented browser layer with:

  • accessibility snapshots instead of huge HTML
  • stable refs for click/type
  • session isolation per user
  • visual login via VNC
  • cookies/storage import
  • screenshots/tracing for debugging
  • transcript extraction and search macros

    • Where it could help OiMy:

  • Vacation planning / booking research
  • Restaurant and activity search
  • Shopping and grocery cart workflows
  • School portal scraping where user authorizes login
  • Insurance/medical-bill portal read-only summaries
  • Form-prep workflows where user confirms final submit

    • Guardrail: Do not position/use it as “bypass Cloudflare/bot detection.” For OiMy, frame it as user-authorized resilient browsing. Do not use it to scrape private data without consent, bypass paywalls, or violate site terms.

    Recommendation: Pilot as an optional browser backend for internal agents only. Do not make it a user-facing feature. Add a policy wrapper: read-only by default, confirm before submit/purchase/send.

    Priority: High pilot value, medium risk.

    ---

    2. pi-mem

    What it is: Plain-markdown persistent memory for coding agents: `MEMORY.md`, `SCRATCHPAD.md`, daily logs, notes, search, and automatic context injection.

    Overlap with OiMy:

    OiMy/OpenClaw already has memory files, daily notes, GBrain, deep profile, and container-scoped workspaces. We do not need to adopt pi-mem wholesale.

    Useful patterns to adopt:

  • Explicit `SCRATCHPAD.md` for open items
  • “Last 24h” session summary at startup
  • daily append-only logs + curated long-term memory separation
  • per-member/private memory + family/shared memory separation
  • autocommit/versioning for memory edits

    • Recommendation: Adopt patterns, not package. Add OiMy memory layers:

  • `PRIVATE_MEMORY.md` per member
  • `FAMILY_MEMORY.md` per household
  • `SHARED_PREFERENCES.md` with explicit consent/source
  • `SCRATCHPAD.md` for unresolved tasks
  • startup “last 24h family brief” for agents

    • Priority: Medium-high, especially for family architecture.

    ---

    3. pi-reflect

    What it is: Reflection engine that reads transcripts and reference files, compares actual behavior against target behavior, and surgically edits memory/behavior files to reduce repeated failures. Tracks correction rate and recidivism.

    Why it matters for OiMy:

    This is directly aligned with Bharath’s complaint about repeated mistakes: overnight GPU jobs left unattended, wrong model verification, missing follow-ups, etc. OiMy needs an automatic “what did we keep getting wrong?” loop.

    Use cases for OiMy:

  • Skill-level reflection: after failures/evals, update SKILL.md pitfalls/tests
  • Agent behavior reflection: update AGENTS.md/HEARTBEAT.md after repeated misses
  • Product QA reflection: trend failures across users/containers
  • User-specific reflection: update deep profile and preferences from accepted/dismissed suggestions
  • Operations reflection: recurring infra mistakes become durable rules

    • Recommendation: Implement an OiMy-native reflection pipeline. Inputs:

  • session transcripts
  • eval reports
  • skill-stats.json
  • HEARTBEAT state
  • user corrections

    • Outputs:

  • proposed skill edits
  • proposed tests
  • memory updates
  • incident notes
  • recidivism metrics

    • Priority: Very high. This is a quality moat and aligns with “test, verify, fix, retest.”

    ---

    4. safe-skill-search

    What it is: Rust local semantic/full-text search across agent skill registries, with quality filtering and scoring.

    Why it matters for OiMy:

    OiMy now has many skills. Discovery is already becoming a UX problem. The skill library search bar helps humans, but agents also need skill discovery with quality and routing boundaries.

    Useful ideas:

  • quality score per skill
  • trusted vs community skills
  • registry search
  • JSON output for agent consumption
  • “only return high quality skills”
  • skill detail pages with examples and install URLs

    • Recommendation: Don’t necessarily install this binary. Implement the pattern in OiMy:

  • quality score in `skill-stats.json`
  • failure rate + eval pass rate + user rating
  • search over skill descriptions/examples/triggers
  • “suggest skill” capability in app/agent
  • skills library sort by quality/newness/category

    • Priority: Medium-high.

    ---

    5. wa_meow

    What it is: Self-hosted WhatsApp bridge using whatsmeow; REST + SSE; multi-session; send messages/audio/docs/reactions/typing; session persistence.

    Why it matters:

    OiMy currently has Twilio WhatsApp infrastructure. Twilio is reliable/compliant but constrained by templates, costs, and Meta approval flows. A linked-device bridge may feel more native and support richer personal-assistant interactions.

    Pros for OiMy:

  • self-hosted
  • less Twilio template friction
  • richer events and reactions
  • potentially better family-group presence
  • multi-account support

    • Risks:

  • WhatsApp linked-device session fragility
  • operational support burden
  • potential ToS/ban risk
  • harder enterprise/compliance story
  • user trust issue if we ask them to link WhatsApp directly

    • Recommendation: Keep as a future experiment, not core production. Use for internal/dev or opt-in power users only. Twilio remains safer for production.

    Priority: Medium, later.

    ---

    6. jo-inc skills / skills-db

    What they are: Official jo skill registry and skill database.

    Useful for OiMy:

    Reference material only. Could compare skill schemas/triggers for gaps.

    Recommendation: Use as a competitor/registry input for quarterly skill gap scans. Not a direct dependency.

    Priority: Low.

    ---

    What this changes in the askjo.ai plan

    The askjo.ai site showed the product pattern: anticipatory daily rhythm + private local/cloud architecture.

    The jo-inc repos show the engineering pattern:

    1. Private local files are durable and inspectable — OiMy should emphasize file-backed memory and household/member boundaries.

    2. Reflection is a system feature — OiMy should automatically learn from failures, not rely on ad hoc human correction.

    3. Browser automation is product infrastructure — serious family assistant workflows need resilient browsing, traces, and VNC login.

    4. Skill discovery needs quality scoring — a growing skills platform needs search/ranking, not just a static list.

    5. Multi-channel is a bridge layer — WhatsApp/Telegram/email are not just notifications; they are user-presence surfaces.

    ---

    Concrete additions to OiMy roadmap

    Add to Wave 1

    1. Reflection QA Loop

    - Nightly/weekly job that reads eval reports + user corrections and proposes skill/memory updates.

    - First target: Food Intelligence and Daily Rhythm skills.

    2. Skill Quality Search

    - Extend skills library + agent skill resolver with quality score, eval pass/fail, user rating, and search.

    3. Private/Shared Memory Split

    - `PRIVATE_MEMORY.md` per member, `FAMILY_MEMORY.md` per household, shared preferences with consent source.

    Add to Wave 2

    4. Resilient Browser Backend Pilot

    - Evaluate camofox-browser for user-authorized workflows: school portals, travel, shopping, forms.

    - Strict “confirm before submit/send/purchase” guardrail.

    5. Last-24h / Daily Context Summary

    - Use pi-mem pattern: at agent start/morning brief, show what happened in last 24h and open scratchpad items.

    Add to Wave 3

    6. Self-hosted WhatsApp Bridge Experiment

    - Evaluate wa_meow for richer WhatsApp presence if Twilio constraints block user experience.

    - Do not replace Twilio without reliability/ToS analysis.

    ---

    Recommended priority

    1. Implement OiMy Reflection QA Loop — highest leverage, prevents repeated mistakes.

    2. Implement household/member memory boundaries — required before deeper askjo-style family coordination.

    3. Add skill quality search/discovery — improves skills platform as it grows.

    4. Pilot camofox-browser for a narrow workflow — school portal or travel planning.

    5. Evaluate wa_meow later, only if Twilio UX becomes a blocker.

    ---

    Bottom line

    The most useful thing from jo is not a repo; it is an operating philosophy:

    > Local/private memory + reflection + reliable browsing + multi-channel presence + proactive daily rhythm.

    OiMy already has many of the product pieces. To surpass jo, OiMy needs the substrate:

  • clean household/member memory boundaries,
  • proactive daily rhythm,
  • reflection-driven quality improvement,
  • skill quality/discovery,
  • and resilient, user-authorized browsing.