[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-85156":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":9,"language":10,"languages":9,"totalLinesOfCode":9,"stars":11,"forks":12,"watchers":11,"openIssues":13,"contributorsCount":14,"subscribersCount":14,"size":14,"stars1d":14,"stars7d":14,"stars30d":14,"stars90d":14,"forks30d":14,"starsTrendScore":14,"compositeScore":15,"rankGlobal":9,"rankLanguage":9,"license":16,"archived":17,"fork":17,"defaultBranch":18,"hasWiki":17,"hasPages":19,"topics":20,"createdAt":9,"pushedAt":9,"updatedAt":25,"readmeContent":26,"aiSummary":9,"trendingCount":14,"starSnapshotCount":14,"syncStatus":27,"lastSyncTime":28,"discoverSource":29},85156,"tomorrowedge","axobase001\u002Ftomorrowedge","axobase001","Local GUI client for full-access multi-model coding agents",null,"TypeScript",56,6,19,0,38.54,"MIT License",false,"master",true,[21,22,23,24],"agent-orchestration","coding-agent","gui-client","multi-model","2026-06-15 10:04:59","# TomorrowEdge\n\n[![CI](https:\u002F\u002Fgithub.com\u002Faxobase001\u002Ftomorrowedge\u002Factions\u002Fworkflows\u002Fci.yml\u002Fbadge.svg)](https:\u002F\u002Fgithub.com\u002Faxobase001\u002Ftomorrowedge\u002Factions\u002Fworkflows\u002Fci.yml)\n\n**English** | [中文](README.zh-CN.md)\n\nTomorrowEdge is a **local governance and policy-evolution runtime for heterogeneous coding agents**.\n\nIt gives engineering teams a GUI\u002Fruntime layer for strong-agent governance, budget-bounded multi-model execution, Agent Council planning, delegated TaskGraph execution, traceable evidence, and offline policy evolution.\n\nIt is not another chat bot, single-model CLI wrapper, benchmark dashboard, or general personal-agent OS. TomorrowEdge turns Codex, Claude Code, DeepSeek, MiMo, Ollama, OpenRouter models, local models, command agents, MCP agents, and custom adapters into replaceable capability nodes inside a governed software-engineering council.\n\n```text\nGoverned orchestration. Full traceability.\nStrong agents should decide. Efficient agents should execute.\nCodex and Claude Code give agents full access.\nTomorrowEdge gives heterogeneous full-access agents governance, trace, budgets, and policy evolution.\n```\n\n![TomorrowEdge Sirius council runtime](docs\u002Fui\u002Fscreenshots\u002Fgui-v1.5\u002Fcouncil-main.png)\n\n## Why It Exists\n\nThe core problem in AI coding is no longer only model strength. Strong agents can already write code.\n\nThe hard part is orchestration and governance:\n\n- Which agent should own the architecture plan?\n- When is a strong-agent call worth the cost?\n- Which cheaper or local agent can do implementation labor safely?\n- Who reviews, judges, and resolves disagreements?\n- What evidence is required before patch or shell execution?\n- How should the system recover when a delegated task fails?\n- How do full-access agents stay visible, auditable, and budget-bounded?\n\nOpenRouter routes requests. TomorrowEdge routes objectives, capabilities, roles, tools, budgets, evidence, task ownership, strategy mutation, and engineering delivery.\n\n## Sirius 1.5\n\n**Sirius** is the TomorrowEdge 1.5 release line. Its main runtime is the **Agent Council Governance Runtime**.\n\nA high-level engineering task enters a Chief Agent first. The Chief Agent can convene Council Members for critique, gap fill, alternative planning, and task claims. The council forms a consensus TaskGraph. Each core task node receives an owner agent, provider, model, and assignment reason. TomorrowEdge then delegates execution while EvidenceGate, BudgetGate, Debate v2, Objective Contract, Strategy Memory, and the event ledger govern the run. Final delivery returns to the Chief Agent for review and judge.\n\n```mermaid\nflowchart LR\n  User[\"High-level task\"] --> Chief[\"Chief Agent\"]\n  Chief --> Council[\"Agent Council\"]\n  Council --> Graph[\"Consensus TaskGraph\"]\n  Graph --> Own[\"Task ownership assignment\"]\n  Own --> Exec[\"Delegated execution\"]\n  Exec --> Gates[\"EvidenceGate + BudgetGate + Debate v2\"]\n  Gates --> Mutate[\"Bounded strategy mutation\"]\n  Mutate --> Final[\"Chief final review \u002F judge\"]\n  Final --> Deliver[\"Deliverable + trace\"]\n```\n\nCore Sirius modules:\n\n- **Chief Agent Router** sends high-level engineering goals to a chief agent first.\n- **AgentCapabilityProfile** makes every model or external agent replaceable by capability, role, trust, cost, latency, and adapter support.\n- **Agent Council Planning** records critique, gap fill, alternative planning, task claims, and consensus moves.\n- **Task Ownership Assignment** gives every core TaskGraph node an `ownerAgentId`, `assignedProvider`, `assignedModel`, and `assignmentReason`.\n- **Delegated Execution Runtime** executes owned nodes while preserving Objective Contract, TaskGraph, RoleGraph, EvidenceGate, BudgetGate, Debate v2, Strategy Memory, and Trace Ledger.\n- **Bounded Strategy Mutation** can split tasks, switch owner agents, add reviewers\u002Fjudges, increase debate, or trigger council replan without mutating safety boundaries.\n- **Chief Final Review \u002F Judge** returns the deliverable to the Chief Agent before completion.\n\nSirius is experimental, but it is now the main product direction.\n\n## Runtime Screenshots\n\nThese are actual runtime screenshots captured from the local `tedge client` GUI against a Sirius council fixture session. They are not static design boards.\n\n### Council Cockpit\n\n![Sirius Agent Council cockpit](docs\u002Fui\u002Fscreenshots\u002Fgui-v1.5\u002Fcouncil-main.png)\n\n### Objective Contract And Trace Drawer\n\n![Sirius details drawer](docs\u002Fui\u002Fscreenshots\u002Fgui-v1.5\u002Fcouncil-details.png)\n\n### API Key And Provider Setup\n\n![API key manager](docs\u002Fui\u002Fscreenshots\u002Fgui-v1.5\u002Fkey-role-manager.png)\n\n### Role Assignment\n\n![Role assignment](docs\u002Fui\u002Fscreenshots\u002Fgui-v1.5\u002Frole-assignment.png)\n\n## What Makes It Different\n\n| Category | Common approach | TomorrowEdge |\n| --- | --- | --- |\n| Single-agent coding CLI | One strong model owns the whole run | Multiple agents are governed by role, capability, cost, trust, and evidence |\n| Model router | Pick a model for one request | Route objectives, roles, tools, context, ownership, verification, and delivery |\n| Agent framework | Build agent graphs | Act as a cockpit and governance layer over native workflows and external frameworks |\n| Benchmark dashboard | Optimize leaderboard outputs | Treat benchmark and dashboard utilities as secondary evaluation tools |\n| Prompt optimization | Tune prompts or fixed workflows | Evolve orchestration policy from objective-action-feedback traces |\n| Full-access tools | Give agents broad power | Keep full-access execution visible, logged, budgeted, reviewable, and reversible |\n\n## Quick Start\n\n```bash\nnpm install\nnpm run build\nnpm run client\n```\n\nThe GUI client opens a local cockpit for natural-language tasks, access-mode selection, provider setup, role routing, approvals, telemetry, details, and trace inspection.\n\nFor installed builds:\n\n```bash\ntedge client\ntedge desktop\n```\n\nRun without API keys:\n\n```bash\nnpm run dev -- run \"fix failing test\" --headless --fixture-mode --approve-patch --approve-shell\n```\n\nRun the Sirius council path:\n\n```bash\nnpm run dev -- council run \"Rewrite a small TypeScript utility as a Rust module with tests\" --headless --fixture-mode\nnpm run dev -- council run \"Rewrite a small TypeScript utility as a Rust module with tests\" --config examples\u002Fconfigs\u002Fsirius-codex-deepseek-mimo.mock.yaml --headless --fixture-mode\n```\n\nOr through the normal `run` command:\n\n```bash\ntedge run \"rewrite this service with a safer architecture\" --agent-council\n```\n\n## Configure Providers\n\nOpen the GUI and click **Keys** to configure providers and role routing. TomorrowEdge stores local secrets through the encrypted local secret store or environment-variable indirection; config files keep env-var references instead of raw keys.\n\nRecommended first setup:\n\n1. Start with OpenRouter if you are not sure. One key gives access to many model families.\n2. Use separate provider keys when possible for rate-limit isolation, cost tracing, and failure diagnosis.\n3. Assign stronger agents to chief\u002Fplanner\u002Freviewer\u002Fjudge roles.\n4. Assign cheaper or local agents to explorer\u002Fcoder\u002Ftest\u002Fdocumentation roles where risk allows.\n5. Keep `auto` when you want TomorrowEdge to choose by routing policy.\n\nExample role intent:\n\n```yaml\nchief_agent:\n  prefer: [strong_reasoning, external_codex, external_claude]\n\nrole_routing:\n  planner:\n    prefer: [strong_reasoning]\n  coder_a:\n    prefer: [coding, fast]\n  coder_b:\n    prefer: [cheap, coding]\n  reviewer:\n    prefer: [strong_reasoning, conservative]\n  judge:\n    prefer: [strong_reasoning]\n```\n\nThe model names in screenshots are examples, not hardcoded product requirements.\n\n## Core Concepts\n\n### Objective Contract\n\nTomorrowEdge inserts a verifiable contract before planning. It defines success criteria, failure criteria, allowed tools, forbidden actions, required evidence, budget bounds, role permissions, and stop conditions. Planner output can add operational detail, but it cannot relax the contract.\n\n### Agent Council\n\nThe Chief Agent can convene a council of replaceable agents. Council members produce structured moves:\n\n- critique\n- gap fill\n- alternative plan\n- task claim\n- consensus revision\n- final consensus\n\n### Task Ownership\n\nTaskGraph nodes are not abstract labels only. Core nodes carry concrete ownership:\n\n- `ownerAgentId`\n- `assignedProvider`\n- `assignedModel`\n- `assignmentReason`\n- fallback candidates\n- evidence refs\n- artifact refs\n\n### Evidence And Trace\n\nRuntime artifacts are preserved for replay, while compact evidence packets are projected to models. Every important action writes to the event ledger: route decisions, budget decisions, model calls, council moves, task results, reviews, judge decisions, patch actions, shell runs, repairs, mutations, and final summaries.\n\n### Orchestration Policy Genome\n\nInspired by evolutionary algorithms, TomorrowEdge makes orchestration policy the unit of evolution. The unit of evolution is not answer, prompt, or agent; it is orchestration policy.\n\nThe system does not evolve model weights, raw prompts, or individual answers. It evolves bounded runtime policies that decide how contracts are generated, how plans are derived, how roles are routed, how evidence is verified, how failures are repaired, how traces are retrieved, and when a run should stop or ask the user.\n\n> The unit of evolution is not the answer, the prompt, or the agent. It is the orchestration policy.\n\nSafety boundaries cannot be mutated.\n\n## Access Modes\n\n| Mode | Behavior |\n| --- | --- |\n| `restricted` | No patch or shell execution. Read-only and advisory workflows only. |\n| `partial` | Patch and shell actions require approval. |\n| `full` | Patch, shell, and repair can run automatically when policy permits. Every step is still logged. |\n\nFull access means governed autonomy plus full traceability, not silent execution.\n\n## CLI Map\n\n```bash\ntedge client                         # local GUI client\ntedge desktop                        # optional local desktop window\ntedge run \"fix failing test\"          # native workflow\ntedge run \"...\" --agent-council       # Sirius council path\ntedge council run \"...\"               # explicit council runtime\ntedge models --connection-test        # provider connectivity checks\ntedge trace latest --verbose          # inspect latest event ledger\ntedge sessions inspect latest         # structured session inspector\ntedge policy inspect                  # policy genome\ntedge policy evolve --offline         # offline policy evolution\ntedge skills list                     # governed skills and tool packs\ntedge mcp serve                       # MCP bridge for external agents\n```\n\nUseful development checks:\n\n```bash\nnpm run build\nnpm run web:build\nnpm run docs:status\nnpm run secrets:scan\nnpm run test:council\n```\n\n## External Agents\n\nTomorrowEdge does not replace Codex, Claude Code, or other coding agents. It can bind them into workflow roles through MCP or command adapters.\n\nExamples:\n\n- Codex as Chief Agent, planner, reviewer, or judge.\n- Claude Code as architecture reviewer or final judge.\n- DeepSeek \u002F Kimi \u002F Qwen \u002F MiMo as implementation, exploration, or test-planning agents.\n- Ollama or local agents for privacy-sensitive roles.\n\nExternal agents should submit typed role outputs instead of opaque final answers.\n\n## Capability Stitching\n\nTomorrowEdge routes capabilities, not just requests.\n\nExample:\n\n```text\nScreenshot \u002F diagram \u002F error image\n  -> Vision Agent\n  -> Structured Visual Spec\n  -> Planner \u002F Coder\n  -> Patch \u002F Test\n  -> Reviewer \u002F Judge\n```\n\nA model that sees images does not need to be the model that writes code. A model that writes fast does not need to be the model that judges risk.\n\n## Documentation\n\n- [Agent Council Governance Runtime](docs\u002FAGENT_COUNCIL_GOVERNANCE.md)\n- [Agent Capability Profiles](docs\u002FAGENT_CAPABILITY_PROFILES.md)\n- [Chief Agent Runtime](docs\u002FCHIEF_AGENT_RUNTIME.md)\n- [Delegated Execution Runtime](docs\u002FDELEGATED_EXECUTION_RUNTIME.md)\n- [Policy Evolution Runtime](docs\u002FPOLICY_EVOLUTION_RUNTIME.md)\n- [Objective Contracts](docs\u002FOBJECTIVE_CONTRACTS.md)\n- [Trace Memory](docs\u002FTRACE_MEMORY.md)\n- [Adaptive Orchestration Runtime](docs\u002FADAPTIVE_ORCHESTRATION.md)\n- [MCP Agent Bridge](docs\u002FMCP_AGENT_BRIDGE.md)\n- [Local Cockpit API](docs\u002FLOCAL_COCKPIT_API.md)\n- [Capability Status](docs\u002FCAPABILITY_STATUS.md)\n- [README Promise Map](docs\u002FREADME_PROMISE_MAP.md)\n\n## Current Status\n\nCurrent version: `1.5.1`.\n\nRelease line: Sirius.\n\nImplemented mainline pieces include:\n\n- GUI client and optional desktop shell.\n- Provider onboarding and connection tests.\n- Role routing and configurable role assignment.\n- Objective Contract.\n- Agent Council Governance Runtime.\n- Task ownership assignment.\n- Delegated execution events.\n- Evidence packets and artifact-aware trace export.\n- Budget governance and real\u002Fsimulated strong-agent telemetry.\n- Debate v2, reviewer, judge, repair, and bounded mutation events.\n- MCP bridge and command external-agent runner skeleton.\n- Offline fixture workflows and deterministic council demo path.\n\nSome parts remain experimental:\n\n- Deep external Codex \u002F Claude Code process integration.\n- Fully asynchronous graph runtime.\n- Long-horizon policy evolution quality.\n- Public comparative evaluation.\n\n## Privacy And Safety\n\nTomorrowEdge is local-first. Sessions, artifacts, traces, and local secrets stay in the local workspace unless you explicitly configure external providers or publish artifacts.\n\nNever commit real API keys. Use environment variables or the local encrypted secret store. The release secret scanner excludes generated artifacts and binary screenshots, but provider keys should still be rotated if they were ever pasted into chat, screenshots, or public logs.\n\n## License\n\nMIT\n",2,"2026-06-15 02:30:10","CREATED_QUERY"]