Roadmap

"People don't know what they want until you show it to them."

This isn't a feature list. It's a manifesto about how your AI assistant should exist.

---

What we believe

Everyone deserves an AI assistant that actually understands them.

Not a chat toy. Not a tech demo. A digital counterpart — one that knows how you work, connects to all your tools, thinks for you, executes for you, and remembers for you.

MateClaw is that thing.

---

What we've shipped

v1.0 — It thinks and acts ✅ Released

Make an AI assistant a coworker who uses tools, not a chat box.

• ReAct engine: reason, act, observe, reason again
• Plan-and-Execute orchestration: plan first, then execute step by step
• StateGraph architecture: state-graph-based agent orchestration
• DynamicAgent: loaded from database at runtime, adjustable without restart
• 20 built-in tools: search / shell / file I/O / delegate / multimodal generation / cron / SQL
• Tool Guard + File Guard + Audit Log: every tool call has approval, control, and a record
• SKILL.md skill system: install new capabilities into your AI like apps

v1.1 — It's everywhere ✅ Released

Move AI out of the chat box on a webpage and into every IM your team actually uses.

• 8 channels: Web / DingTalk / Feishu / WeCom / Telegram / Discord / QQ / WeChat Personal / Slack
• 4-layer memory: session context + workspace memory + post-chat extraction + 2 AM consolidation
• DREAMS.md consolidation diary: human-readable audit of memory changes
• Workspace isolation: every agent / skill / wiki / conversation / memory belongs to a workspace
• ChatGPT OAuth + Anthropic Claude Code OAuth: log in with your subscription, no API key
• LLM Wiki + RAG: raw files become structured pages with bidirectional links and summaries

v1.2 — It's your coworker ✅ Released (2026-05-05)

Renamed "agents" to digital employees — not vocabulary purism, a worldview shift.

• Digital employees with Role / Goal / Backstory — not a cold system prompt
• 5 career templates: product researcher / customer support / knowledge curator / data analyst / executive assistant
• Skills are backbones: each skill has its own SKILL.md + LESSONS.md + workspace filesystem
• ACP bridge: Claude Code, Codex, Gemini CLI plug in as employees
• Backstage runtime console: for the first time you can see what each employee is doing right now
• Onboarding wizard + Dashboard + Doctor

Full story: v1.2.0 release notes.

v1.3 — It orchestrates business flows ✅ Released (2026-05-13)

Graduating from a chatbot framework to a business-process OS — a flow is no longer several employees chatting separately, but a publishable, triggerable, replayable linear-step DSL.

• Workflow: 7 step modes (sequential / fan_out / collect / conditional / await_approval / dispatch_channel / write_memory) + Pebble expressions + JSON-first authoring + integer revisions + run history
• Natural language → workflow draft: describe the flow, an agent emits graph_json, a human reviews before publish
• Triggers: 6 pattern types (cron / webhook / channel_message / agent_lifecycle / content_match / workflow_completion), event governance on by default (dedup / rate limit / recursion guard)
• Persistent await_approval pause: survives service restarts
• Image editing, 4 document-generation tools (Docx/Xlsx/Pptx/Pdf), MCP per-agent tool binding, multimodal sidecar routing

Full story: v1.3.0 release notes.

v1.4 — It's more autonomous and leads teams ✅ Released (2026-05-23)

Flows were scripted by you, but the employee itself still "answered one round and stopped." This release puts the focus back on the employee.

• Persistent goals: say it once — the employee locks the goal, self-checks every round, and keeps itself going until done or out of budget
• Sub-employee delegation tree: recursive delegation up to 3 levels deep, with sync / parallel fan-out / async delegation tools; the Employee Builder spins up a whole team from one sentence
• Progressive tool/skill disclosure: core tier always visible, extension tier activated on demand via enable_tool / load_skill — pile on tools without blowing the context
• Workspace RBAC: Owner / Admin / Member / Viewer roles + capability gates — MateClaw is usable by a team for the first time
• Feishu as a first-class citizen: interactive cards, approval cards, streaming cards, voice transcription, file/audio/video I/O, channel-native tools
• Native Gemini, xAI / Grok, per-conversation model pinning, structured context compaction, rate-limit failover

Full story: v1.4.0 release notes.

v1.5 — It's verifiable, knowledge self-maintains, memory knows its owner ✅ Released (2026-06-04)

Make autonomy verifiable, knowledge self-maintaining, and memory owner-aware.

• Goal checklists: goals decompose into independently verifiable criteria; the evaluator checks them off one by one — all checked or it's not done. No "95% is close enough"
• Self-maintaining Wiki: [[wikilink]] page interlinking + rename/delete cascade rewrites + broken-link lint; fact vs. experience knowledge layers with staleness propagation; pageType profiles + per-agent permissions; event-triggered processing pipelines; local directories mounted as knowledge sources with incremental sync
• Per-owner memory isolation: every memory carries an owner_key and visibility scope (personal / team / global) — one employee serves a whole group without cross-talk; APIs pass through endUserId
• Primary KB per employee, preferred-provider routing that actually applies, generated files persisted to disk

Full story: v1.5.0 release notes.

v1.6 — It meets you where you are ✅ Released (2026-06-22)

Where it can run, what it can do with hands and eyes, and how directly you shape who it is.

• KingbaseES + PostgreSQL as first-class citizens: the PostgreSQL family shares one migration tree; regulated / domestic-procurement environments covered; MySQL and desktop H2 untouched
• Images persist across turns: the screenshot you sent three messages ago is still visible on follow-up; image_analyze re-reads on demand
• execute_code: the employee writes code and runs it — arithmetic, file conversion, verification become real actions instead of guesses
• Shape the employee's identity: a real editor for AGENTS.md and other context files (modal + section reorder); an About You identity block; the employee knows which model it runs on
• Scoped KB access + Wiki Sources tab (multi-path + glob + per-KB auto-sync)
• Global outbound proxy, deterministic Markdown normalization of final answers

Full story: v1.6.0 release notes.

v1.7 — It's ready for production ✅ Released (2026-07-04)

A productionization pass: once you put it into real collaboration, the places that go invisible, un-closable, out of reach, oversized for the window, and walled off — all fixed.

• All three approval paths closed end-to-end: workflow await_approval actually pushes to channels and resolves → resumes execution; the WebChat (API-key) channel can approve/deny and replay; Feishu/WeCom card buttons directly resolve workflow approvals
• Long tasks are visible: an always-on Run Overview rail (step progress + live delegated sub-agent tree) + a per-turn token breakdown (cache hit/miss/write + reasoning split) + sub-agent cost rolled up + one-click generated-file download
• Fits the real model window: local-model context-window probing, a unified token budget for prefix injection, small-context degradation, and tool-schema budget gating — no more "guess 32K" pre-flight rejections or silent truncation
• Opens up: a knowledge-base + Deep Research open API (API-key + rate limit + SSE), a pluggable search Provider SPI, and MCP identity forwarding (carry the authenticated user's identity into a STDIO MCP)
• Reaches further: desktop local-embedded / remote-centralized dual mode + multi-server switching + the mateclaw-desktop source opened; a LAN deployment mode opens controlled intranet access
• One-click operational data export: 9-sheet Excel from the Dashboard + a CLI for offline export
• Wiki processing-failure visibility, per-employee model chains, OpenAPI / Swagger directly debuggable, chat back-to-bottom floating button

Full story: v1.7.0 release notes.

v1.8 — It does a whole job ✅ Released (2026-07-12)

The employee turns outward and finishes a whole job — from a one-sentence brief to a publishable post — on MateClaw's own primitives.

• Content Studio — the first flagship scene: a seeded employee runs pick-topic → research → draft → illustrate → de-AI → layout → deliver. WeChat Official Account (公众号) image-text articles (inline-style HTML → draft box) and Xiaohongshu (小红书) image-first notes (≥3 vertical 3:4 cards + online preview) ship first-class
• De-AI-ification you can measure — a heuristic AI-trace score drives a detect → rewrite → re-check loop, capped at 3 rounds
• A publish chain hardened for real operation — body images uploaded into WeChat, AES-GCM-encrypted secrets, reused service + persisted token, retry + Chinese error hints, guaranteed fallback cover; draft-box-first, publish approval-gated
• A content calendar that dedups and remembers — every delivery is compliance-scanned and auto-recorded, a topic fingerprint stops repeat picks, and a read-only Content Calendar page shows drafted/packaged/published/failed
• The browser agent sees by reference — an accessibility-tree ref snapshot + interact-by-ref, real-browser privacy guardrails, and a controlled CDP escape hatch
• Attention anchoring & environment awareness, a tool-call loop guard, a post-mutation verify reminder; a fast-load pass (~78% smaller initial bundle), a context-occupancy panel, cross-KB wikilinks, MCP progress notifications, a Volcano Engine provider, PostgreSQL 16

Full story: v1.8.0 release notes.

v2.0 — It leads a team ✅ Released (2026-07-26)

From "one person who gets things done" to "a team that collaborates" — Agent Teams become a standing roster around a shared task board.

• Team entity and roles: a team = name + lead + members + reviewers — persisted, reusable; the roster and collaboration playbook inject into member prompts
• Shared task board: eight-status kanban, blockedBy dependency orchestration, member-level parallel dispatch, automatic prerequisite hand-off, settled results waking the lead
• Lead dispatch: the lead decomposes, assigns, reviews; a Plan-Execute lead hands its whole plan over to the board — steps become tasks, dependencies become parallelism
• Execution hardening: lease heartbeats against double execution, cancel-interrupt, in_review human approval gates, retry for failed/stale
• Deliverables and full observability: output files register on tasks, task timelines, a team SSE live board, and jump-in access to any member's word-by-word run
• Plus: workspace isolation fully sealed, channel magic commands + the WeCom progress bubble, conversation rewind/regenerate, explainable auto-approval, policy-driven LLM error recovery, in-chat attachment preview, single-source SKILL.md, the Mem0 plugin provider

Full story: v2.0.0 release notes; user guide: Agent Teams.

---

Next: Agent Loop & Team follow-through

"Great things in business are never done by one person. They're done by a team of people."

Look back along the line: v1.2 gave employees an identity, v1.3 made flows orchestratable, v1.4 made employees follow goals and spin up delegation trees, v1.7 made long tasks visible, and v2.0 made teams a standing roster.

One "stop" remains: employees are reactive. Goal auto-followup only lives within a single run; cron and triggers can wake an employee up, but every wake-up is an isolated response. No employee is truly on duty — continuously watching its area of responsibility and deciding for itself when to act.

Agent Team follow-through — the roster exists; now it grows skills

2.0 delivered the team entity, the task board, lead dispatch and the hardened execution chain (above). Still on the team track:

• [ ] Peer review: critical deliverables can require another member's sign-off before shipping (2.0's in_review is human approval; member peer review is the next step)
• [ ] Team-level goals: one goal decomposes into member sub-goals; the checklist aggregates across members — hover the lead's avatar to see what the whole team still owes
• [ ] Team-to-channel binding: bind a Feishu / DingTalk group to a team; @ the team in the group, the lead decides who takes it
• [ ] Team retrospectives: task wrap-up auto-generates a retrospective into the team's LESSONS.md — this team does better next time
• [ ] Collaboration DAG / swimlane view: draw task dependencies and member swimlanes on top of the timeline data
• [ ] Employee Builder upgrade: one sentence emits a standing team with a roster

Agent Loop — from "answers then stops" to "on duty"

A new state for employees: on duty. Not waiting for you to speak, but cycling autonomously on a heartbeat — wake → check inbox and goals → decide whether to act → act → journal → sleep:

• [ ] Resident loop runtime: an employee can be set "on duty," waking on a configurable heartbeat (minutes to days) to check its area of responsibility
• [ ] Task inbox: channel messages, trigger events, delegations from other employees, to-dos you toss over — one queue, consumed by priority on each wake-up
• [ ] Cross-session goal continuation: v1.4/v1.5 auto-followup lives inside a single run; the loop carries goals across sessions and across days until every criterion is checked
• [ ] Budgets and circuit breakers: per-loop token / cost / turn budgets; consecutive failures trip the breaker into sleep pending your decision; ToolGuard approval gates still intercept sensitive actions — autonomy is not loss of control
• [ ] Loop journal: what it did each wake-up, why it chose not to act, what it spent — human-readable and replayable, what DREAMS.md is to memory
• [ ] Pause / resume / clock-out: controllable from the UI and from channel commands; the Run Overview sidebar shows every on-duty employee's loop state
• [ ] Quiet hours and interruption policy: silent accumulation at night, proactive reporting for what matters — integrated with the nudge system, it knows what's worth waking you for

Where they converge: a department that runs itself

A leader on a loop, members summoned on demand — that's a self-running digital department:

• Morning-report department: the leader wakes at 7:00, dispatches data collection, analysis, and writing to members, peer-reviews, posts to the group — you wake up to results
• Support department: a ticket lands in the inbox, the leader classifies, assigns the right member, escalates to you what it can't handle
• Intelligence department: a monitoring employee loops over sources, wakes the analyst only when something changed, notifies you only when it's worth interrupting

Workflows own the deterministic processes; teams + loops own the unpredictable everyday. They complement each other — none replaces another.

Advancing in parallel

• [ ] Workflow loop / invoke_skill step modes: per-item array iteration / call a skill without going through an employee
• [ ] Workflow canvas editing: from read-only chain rendering to drag-to-edit
• [ ] Run replay view: trace timeline + input/output diff on any node
• [ ] Scenario templates and marketplace: package "employees + team + workflow + triggers + KB structure" into one-click-importable scenario bundles

---

What we deliberately don't do

"I'm as proud of the things we haven't done as the things we have done."

Cut	Why	When it might return
Fine-grained RBAC beyond four roles	v1.4's Owner / Admin / Member / Viewer + capability gates cover real team needs. Button-level permissions and custom role composition belong to enterprise management platforms	When real multi-team SaaS customers need fine-grained permissions
Multi-tenancy	Premature multi-tenancy is architectural cancer. Workspace isolation already covers multiple teams in one org	When there's a clear SaaS commercialization path
SSO / LDAP / SAML	Enterprise integration is a bottomless pit	When paying enterprise customers explicitly ask
30+ node visual workflow editor	7 step modes already cover 90% of real-world scenarios; the rest is pushed to natural-language generation	When a user case actually needs 30+ nodes (rare)
Native mobile app	8 IM channels + desktop (now with remote connect) + Web already cover it. On your phone, you use MateClaw via DingTalk / Feishu / Telegram	When Web / IM channels can't deliver an irreplaceable mobile-only feature
Replacing ReAct / Plan-Execute	Workflows, teams, and loops collaborate with those two engines, not replace them — single-agent multi-turn reasoning still lives there	Never replaces
Unbudgeted full autonomy	Agent Loop always ships with budgets, circuit breakers, and approval gates. "Run until the money's gone" isn't autonomy, it's loss of control	Never

---

Version milestones

Version	One line	User experience goal	Status
v1.0	It thinks and acts	An AI assistant that uses tools to solve problems	✅ Released
v1.1	It's everywhere	8 channels + 4-layer memory + workspaces + LLM Wiki	✅ Released
v1.2	It's your coworker	Digital employees + career templates + backbone skills + ACP bridge + Backstage	✅ Released
v1.3	It orchestrates business flows	Workflow + triggers + document generation + per-agent tool binding	✅ Released
v1.4	It's more autonomous and leads teams	Persistent goals + delegation tree + progressive disclosure + RBAC + first-class Feishu	✅ Released
v1.5	It's verifiable	Goal checklists + self-maintaining Wiki + owner-aware memory	✅ Released
v1.6	It meets you where you are	Domestic databases + persistent vision + code execution + identity shaping	✅ Released
v1.7	It's ready for production	Approval paths closed + Run Overview & cost visibility + context/token budgeting + open API/Deep Research + desktop remote/LAN + operational export	✅ Released
v1.8	It does a whole job	Content Studio — one sentence to a publishable 公众号 / 小红书 post + browser ref interaction	✅ Released
v2.0	It leads a team	Agent Teams + a shared task board — the lead decomposes and dispatches, members run in parallel, deliverables and full observability	✅ Released
Next	It's on duty	Agent Loop resident cycles + team follow-through (peer review / team goals / group binding / retrospectives) = a department that runs itself	📋 Planned

---

One More Thing

We're not building MateClaw to chase ChatGPT, not to be the next Dify, not to add another buzzword to a funding deck.

We're building it because we believe one thing:

AI shouldn't be a chat box on a webpage. It should be your second brain.

It lives in your DingTalk, your Feishu, your Telegram. It's read every document you have. It remembers what you said three months ago. It uses your company's internal tools. It consolidates memory while you sleep. It runs an entire business flow on your behalf. Soon it will lead a standing team, stay on duty, and watch over the things you can't get to.

Someday, you'll forget it's a program.

That's the day we win.

---

Stay hungry. Stay foolish.