20:"$Sreact.view_transition" 21:I[9754,["/_next/static/chunks/2t36f5p4pjzlb.js","/_next/static/chunks/1gyqeidjm5eoa.js","/_next/static/chunks/0_fjnxc4tt_l_.js","/_next/static/chunks/1ymt1shyhmu-4.js","/_next/static/chunks/3kv2zf0i7s2wc.js","/_next/static/chunks/1vs_3g_g75zmk.js","/_next/static/chunks/2g4dh802s8oz3.js","/_next/static/chunks/1bc3f0shukf1j.js","/_next/static/chunks/192kgs9v_t14j.js","/_next/static/chunks/0z6pw0v5c_lx5.js","/_next/static/chunks/08r60e6icmtkc.js","/_next/static/chunks/3trx6c0i7_x-3.js","/_next/static/chunks/40sro5bzp9bi_.js","/_next/static/chunks/1wgf3tk38y6cf.js"],"ProjectDetail"] 22:T5b96, AgentBridge — 为现有业务系统添加 Agent 能力

AgentBridge // 自托管 · AI 可读 · Vibe Coding

让 Agent 进入你现有的业务系统，却不用改一行代码

AgentBridge 是一个 AI 可读的基础平台：你和 AI 只需写下业务领域的工具与规则，平台负责那些共享而困难的底层行为——JSON/SSE、会话生命周期、权限、审批、审计与 RAG 端口。敏感动作在放行前，一律要经过一个清晰的人工闸门。

语言 Python 许可 MIT 作者 foamtor 状态 v1.0.0 单节点

核心机制 · 审批闸门环线

work_order_ops · 参考环线

平台拥有统一治理 · 人工闸门拦截敏感步骤

查询

AI 通过受控协议向你的业务系统发问，解析并结构化你的领域数据。

知识检索

经由 RAG 端口检索组织内的既有知识，作为判断的上下文。

创建工单

AI 起草业务动作——例如生成一张工单，准备写入或执行。

▲

人工审批

闸门在此拦截：放行前必须由人签核，这是环线上唯一的断点。

[ GATE ] 未放行 → 动作挂起 · 放行 → 继续执行

每个环路都建在共享底座之上：session 生命周期 · perms · approvals · audit · JSON/SSE · RAG ports。

签名视觉由 work_order_ops 参考实现推导而来，非截图。

环线的四段

环线不是为了转圈，而是为了设一个闸

参考实现 work_order_ops 可以完整跑通一个业务回合。它把动作画成一条环形通路，唯独在需要授权的地方硬生生断开——那是留给人的位置。

路径标签

query → retrieval → create → approval · gate → release | deny

monospace 携带协议 / 路径与状态证据，叙述由正文承担。

归属边界

平台替你扛住共享而艰难的底层行为

AgentBridge 有意划清一条线：你和 AI 写业务专属的部分，其余全部交还平台——因为这些行为一旦在多处反复实现，就会出现治理裂缝。

平台拥有 · 托管行为

JSON / SSE双向结构化消息与流式推送
Session 生命周期会话的建立、保持与回收
Permissions访问与动作的授权边界
Approvals敏感动作的人工签核流程
Audit可追溯的动作审计记录
RAG ports接入既有知识的检索端口

你与 AI 所写

≈业务领域模型与数据结构

≈业务工具与动作定义

≈专属规则与判断逻辑

平台不取代你的业务代码，只把你反复重写的困难部分收编为统一底座。

参考工作流

一个完整跑通的业务回合

work_order_ops 是仓库自带的参考实现。从提问到签核，一条清晰的通路串起查询、检索、生成与人工断点。

step 01

QUERY

向业务系统提出请求，解析落入领域容器。

→

step 02

RETRIEVAL

经由 RAG 端口检索知识，取得判断上下文。

→

step 03

CREATE

生成工单，动作准备就绪、等待放行。

→

step 04 · gate

APPROVAL

人工闸门拦截，未签核前动作不会继续。

环线在此收束：放行后动作落地并计入审计，下一回合再次从查询开始。

适用边界

什么时候用它，什么时候别用

AgentBridge 面向需要快速验证 AI 能力的团队——尤其当你已经有 API 或数据库。但它不是银弹，README 也把话说明白了。

适合的场景

✓已有 API 或数据库的系统
在既有接口之上叠加 AI 能力，无需改写。
✓想快速验证 AI 的团队
用平台起步，比从零搭建治理层更快。
✓需要共享治理的多系统环境
把权限、审批、审计集中到一处。

慎用的场景

—生产部署演练未完成
迁移 / 恢复与演练属于后续工作。
—多实例需显式配置
多节点扩展不会自动生效。
—不适用于实时交易
实时撮合等场景不在设计范围内。

本地运行

两条命令，跑起来看

参考实现通过 Docker Compose 启动即可。这是仓库记载的真实动作，不是演示截图。

启动 $ docker compose up --build

跟随日志 $ docker compose logs api

github.com/foamtor/agentbridge status v1.0.0 · 单节点稳定 · MIT · 由 foamtor 维护

23:T5c5f, AgentBridge — Add Agent abilities to your existing business systems

AgentBridge // self-hosted · AI-readable · Vibe Coding

Let an Agent step into your business system — without touching a line of code

AgentBridge is an AI-readable foundation you host yourself. You and the AI write the business-specific domain and tools; the platform owns the hard shared behavior — JSON/SSE, session lifecycle, permissions, approvals, audit, and RAG ports. Before a sensitive action is released, it must clear a single human gate.

Language Python License MIT Author foamtor Status v1.0.0 single-node

Core mechanism · the approval gate loop

work_order_ops · reference loop

one shared governance layer · a human gate interrupts sensitive steps

QUERY

Ask your business system through a governed protocol; answers land structured in your domain.

RETRIEVAL

Pull existing knowledge through a RAG port to ground the judgment.

CREATE

The AI drafts a business action — for instance a work order — ready to write or execute.

▲

APPROVAL

The gate holds here: a human must sign off before release. It is the loop's only stopping point.

Every loop runs on a shared base: session lifecycle · perms · approvals · audit · JSON/SSE · RAG ports.

The signature visual is derived from the work_order_ops reference implementation, not a screenshot.

The four segments

The loop is less about circling than about where it stops

The work_order_ops reference implementation runs a complete business round. It draws the action as a circular path, then deliberately breaks it at the one place authority is required — the seat left for a human.

Path labels

query → retrieval → create → approval · gate → release | deny

Monospace carries protocol, path, and status evidence; the narrative voice lives in the text face.

The ownership boundary

The platform carries the hard, shared bottom layer

AgentBridge draws a deliberate line: you and the AI write what is specific to your business, and everything else returns to the platform — because the moment these behaviors are re-implemented in many places, governance starts to crack.

Owned by the platform · managed behavior

JSON / SSEbidirectional structured messaging and streams
Session lifecyclesession start, hold, and reclamation
Permissionsthe authorization boundary for access and actions
Approvalshuman sign-off flow for sensitive actions
Audittracable record of every action
RAG portsretrieval ports into existing knowledge

What you and the AI write

≈your domain model and data structures

≈your business tools and action definitions

≈your own rules and judgment logic

The platform doesn't replace your business code. It gathers the hard parts you would otherwise rewrite everywhere into one shared base.

Reference workflow

A complete business round, run end to end

work_order_ops is the reference implementation shipped with the repo. From question to sign-off, one clear path threads together query, retrieval, creation, and the human stopping point.

step 01

QUERY

Ask the business system and land replies into the domain container.

→

step 02

RETRIEVAL

Retrieve knowledge through a RAG port to ground the judgment.

→

step 03

CREATE

A work order is drafted, the action standing by for release.

→

step 04 · gate

APPROVAL

The human gate holds — nothing continues until it is signed.

The loop closes here: on release the action lands and is audited; the next round begins again at query.

The fit boundary

When to use it, and when not to

AgentBridge is built for teams that want to validate AI quickly — especially when you already have an API or a database. But it isn't a silver bullet, and the README says so plainly.

A good fit

✓Systems with existing APIs or databases
Lay AI ability on top of interfaces you already have, with no rewrite.
✓Teams that want a quick AI validation
Start on the platform rather than building a governance layer from zero.
✓Multi-system environments needing shared governance
Concentrate permissions, approvals, and audit in one place.

Be careful with

—Production deployment drills pending
Migration, recovery, and drills are still follow-up work.
—Multi-node needs explicit configuration
Scaling to several nodes does not happen on its own.
—Not suited to real-time trading
Real-time matching-style scenarios fall outside the design scope.

Run it locally

Two commands, then see it yourself

The reference implementation boots through Docker Compose. These are real actions recorded in the repo, not a demo screenshot.

Start $ docker compose up --build

Follow the logs $ docker compose logs api

github.com/foamtor/agentbridge status v1.0.0 · single-node stable · MIT · maintained by foamtor

24:T400,Both zh-CN and en pages faithfully implement the selected Direction A 'The Approval Gate Loop' brief: the full 6-section sequence is present, all copy has material basis in repository facts / summary / README, and the evidence-derived radial gate-loop signature visual with an interrupt/gate node genuinely exists (SVG ring in the anatomy section plus the 4-step loop panel in the hero). The shared-behavior boundary, reference workflow, when/when-not caveats, and real docker commands are all sourced from evidence. Bilingual sections are equivalent, natural, and keep canonical protocol terms (JSON/SSE, work_order_ops, docker) per the locale contract. No fake controls or device shells. Two non-blocking warnings: the anatomy radial diagram stays a shrinking ring on mobile rather than literally unswirling into a linear pipeline per the brief's mobileTransformation, and the hero uses a subtle ambient radial-gradient backdrop that only partially stands in the forbidden 'decorative glow' shade. Neither blocks approval.a:[["$","$20",null,{"enter":{"secondary-forward":"secondary-forward","secondary-back":"secondary-back-enter","default":"none"},"exit":{"secondary-forward":"secondary-forward","secondary-back":"secondary-back-exit","default":"none"},"update":{"secondary-forward":"secondary-forward","secondary-back":"secondary-back-update","default":"none"},"default":"none","children":["$","$L21","cmt2zh05c00m36271qj5b1rg5:zh-CN:beautified",{"project":{"id":"cmt2zh05c00m36271qj5b1rg5","slug":"foamtor-agentbridge","name":"agentbridge","description":"自托管的 Vibe Coding 业务工具基础平台","authorName":"foamtor","authorVerification":{"verified":false,"verifiedAt":null},"category":"development_tools","subcategory":"data_ai_tools","experienceUrl":null,"qualityScore":56,"featured":false,"officialSite":null,"views":{"original":{"available":true},"official":{"available":false,"reason":"disabled_by_policy"},"beautified":{"available":true,"reason":null},"defaultView":"beautified"},"beautifiedView":{"revisionId":"cmt2zh05s00mf6271m0dzk57a","kind":"artifact","artifact":{"schemaVersion":"localized-static-html-v4","kind":"static_html","locales":["zh-CN","en"],"defaultLocale":"zh-CN","documents":{"zh-CN":"$22","en":"$23"},"designDirections":{"schemaVersion":"dodot-design-directions-v1","directions":[{"id":"dir-approval-gate-loop","name":"The Approval Gate Loop","projectMetaphor":"AI enters an existing business system through a guarded crossing; every sensitive action must pass a single human gate before it is allowed. The loop is the product's core mechanism.","signatureVisual":"A circular work-order loop (query → knowledge retrieval → work order creation → human approval) drawn as a radial road, with one distinct gate node that interrupts the path wherever approval is required.","heroGeometry":"Radial / concentric annotation: a large ring with an interrupt node and a moving token that stops at the gate, derived from the work_order_ops reference flow.","sectionSequence":["hero","gate-loop-anatomy","shared-behavior-boundary","reference-workflow","when-and-when-not","run-local"],"surfaceModel":"Papered control surfaces: an inset 'gate plate' panel, raised ledger cards, and an explicit break band where the human must sign off; depth set by overlay order, not by uniform shadows.","palette":"Ink on warm desk surface — deep slate inks, a single amber 'gate' signal accent, muted paper white, and an olive stop-tone for the approval node.","typography":"Monospace for protocol/path labels and gate status, a condensed display face for the loop title, and a readable text face for narrative; monospace carries evidence, serif-ish text carries the voice.","desktopComposition":"Split plan: the loop diagram holds the left two-thirds as the hero visual, with the approve/deny gate toggle revealed as the actual interrupt; sections alternate wide-loop and narrow-plate widths to keep rhythm.","mobileTransformation":"The polar loop 'unswirls' into a linear 4-step pipeline with the gate as a full-width approval strip; cards stop being radial and become stacked decision stubs.","mediaStrategy":"No product media exists in the manifest; the derived loop diagram is the signature visual and no screenshot frame is fabricated.","forbiddenPatterns":["generic hero slogan over three metric cards","feature-card wall that could host any project's copy","uniform radius, shadow, and blur across every section","decorative gradients, glows, grids, or particles as ornament","fake tabs, search, switches, inputs, or progress bars"],"compositionFingerprint":"radial-gate-loop:ring-with-interrupt-node"},{"id":"dir-bridge-truss","name":"The Bridge Truss Plane","projectMetaphor":"The platform is a truss that spans the gap between an AI and an existing business system; the shared behaviors are the structural struts that keep the span stable, not the payload the AI carries.","signatureVisual":"A bridge cross-section spanning two banks (AI side and existing-code side), where JSON/SSE, sessions, permissions, approvals, audit, and RAG ports are labeled struts holding the deck.","heroGeometry":"Horizontal-span / triangulated deck: a left-to-right truss whose diagonal braces each carry a shared-behavior label, showing what the platform itself supports.","sectionSequence":["hero","span-loading","deck-of-shared-behavior","carried-payload","reference-workflow","run-local"],"surfaceModel":"Structural surfaces: cable/edge lines as borders, deck plates as section fields, and a cantilevered margin that lets content overhang rather than sit in padded boxes.","palette":"Engineering sheet tones — cool graphite structure, cyan line accents for AI rails, warm timber for the existing-code bank, and a low-saturation safety highlight for approvals.","typography":"Structural grid framing: uppercase, letter-spaced engineering labels for struts; a wide, sturdy display face for the bridge statement; mono tags for protocol endpoints.","desktopComposition":"Variable-width horizontal bands emulate span tension: a wide central deck that narrows at both banks; content loads onto the deck in oblique diagonal rows rather than equal columns.","mobileTransformation":"The full-span truss collapses into a two-bank portrait: AI bank stacked above existing-code bank, with struts reduced to vertical load columns between them.","mediaStrategy":"No qualified local media; the truss diagram is a derived structural illustration and stays diagrammatic, never impersonating a dashboard or browser frame.","forbiddenPatterns":["three-metric-card opening","repeated card wall independent of project copy","browser, phone, terminal, or editor shell frames","uniform surface radius on all sections","glowing particle or gradient fields as the visual idea"],"compositionFingerprint":"truss-deck:label-strut-span-between-two-banks"},{"id":"dir-readable-contract-ledger","name":"The Readable Contract Ledger","projectMetaphor":"AgentBridge makes the system AI-readable: every behavior is a typed, auditable contract the platform commits to, and the page reads like a ledger of those commitments rather than a promise of value.","signatureVisual":"A structured protocol ledger: rows of typed contract entries (session lifecycle, permissions, approvals, audit, RAG ports) rendered like records with clear headers, columns, and a commit line, spelling out the shared behaviors.","heroGeometry":"Tabular / rule-lined composition: ruled columns and a wide header taxonomise the platform's owned behaviors as readable, auditable contracts.","sectionSequence":["hero","contract-schema","owned-behavior-ledger","reference-workflow","when-and-when-not","run-local"],"surfaceModel":"Ledger sheets: ruled underline borders, field-column clusters, an auditable signature strip for approvals, and hand-shade fills showing structured rows of data.","palette":"Archive cool — near-black ink, warm ivory sheet, sepia ruling lines; a single seal-red accent reserved for the approval/audit signature marks.","typography":"Ledger-dominant type: tight tabular serif numerals and ruled headers, monospace for field names and protocol payloads, and a quiet lining text face; no heavy display flourish.","desktopComposition":"Column-ruled registry: hero is a large structured readout; sections present as register sheets with alternating field-column clusters, keeping a strong left rule that anchors navigation.","mobileTransformation":"The ruled register rotates to single-column record rows where each contract entry expands to an individual record card with its header line preserved.","mediaStrategy":"No product screenshot exists; the ledger surface is the signature visual and is rendered purely from evidence without mock controls.","forbiddenPatterns":["marketing hero over a metric strip","grid of interchangeable feature tiles","shell/device frames of any kind","uniform glow or blur dressing across sections","fabricated table data, charts, or gauges passed as real product output"],"compositionFingerprint":"ledger-sheets:ruled-contract-columns-with-signature-strip"}],"selectedDirectionId":"dir-approval-gate-loop","selectionRationale":"Direction A best expresses AgentBridge's actual mechanism: the reference implementation work_order_ops is a full loop (query → knowledge retrieval → work order creation → human approval), and the product's differentiator is precisely that AI can act while a human gate intercepts sensitive steps. This yields an evidence-derived radial signature visual, an anatomy section over the loop, a boundary section for what the platform owns (JSON/SSE, sessions, permissions, approvals, audit, RAG), a reference-workflow section from the README, and a when/run section from the documented caveats and commands. Direction A uses the strongest available evidence (a concrete implemented business loop) rather than a generic promise or an abstract structure/ledger, and its mobile unswirl transformation produces genuinely different spatial decisions. Directions B and C are ruled out because they stay abstract (a structural span / an archival ledger) and would rely on thinner evidence for a fully rendered signature visual."},"designBrief":{"schemaVersion":"dodot-design-brief-v1","selectedDirectionId":"dir-approval-gate-loop","projectMetaphor":"AI enters an existing business system through a guarded crossing; every sensitive action must pass a single human gate before it is allowed. The loop is the product's core mechanism.","signatureVisual":"A circular work-order loop (query → knowledge retrieval → work order creation → human approval) drawn as a radial road, with one distinct gate node that interrupts the path wherever approval is required.","compositionFingerprint":"radial-gate-loop:ring-with-interrupt-node","heroGeometry":"Radial / concentric annotation: a large ring with an interrupt node and a moving token that stops at the gate, derived from the work_order_ops reference flow.","sectionSequence":["hero","gate-loop-anatomy","shared-behavior-boundary","reference-workflow","when-and-when-not","run-local"],"surfaceModel":"Papered control surfaces: an inset 'gate plate' panel, raised ledger cards, and an explicit break band where the human must sign off; depth set by overlay order, not by uniform shadows.","palette":"Ink on warm desk surface — deep slate inks, a single amber 'gate' signal accent, muted paper white, and an olive stop-tone for the approval node.","typography":"Monospace for protocol/path labels and gate status, a condensed display face for the loop title, and a readable text face for narrative; monospace carries evidence, serif-ish text carries the voice.","desktopComposition":"Split plan: the loop diagram holds the left two-thirds as the hero visual, with the approve/deny gate toggle revealed as the actual interrupt; sections alternate wide-loop and narrow-plate widths to keep rhythm.","mobileTransformation":"The polar loop 'unswirls' into a linear 4-step pipeline with the gate as a full-width approval strip; cards stop being radial and become stacked decision stubs.","mediaStrategy":"No product media exists in the manifest; the derived loop diagram is the signature visual and no screenshot frame is fabricated.","evidencePlan":["hero establishes project identity, one-line value frame, primary CTA, and the radial gate-loop signature visual","gate-loop-anatomy annotates the four implemented steps (query, knowledge retrieval, work order creation, human approval) from the work_order_ops reference flow","shared-behavior-boundary lists the platform-owned behaviors (JSON/SSE, session lifecycle, permissions, approvals, audit, RAG ports) and marks what the AI and developer write versus what the platform owns","reference-workflow renders the complete documented business loop with the human approval node as the interrupt","when-and-when-not surfaces the documented caveats (production deployment drills pending, multi-node needs explicit config, not for real-time trading)","run-local provides the documented commands (docker compose up --build, docker compose logs api) as real, evidence-sourced actions"],"localeContract":{"zh-CN":"中文页面与英文页面共享同一简报与章节顺序；文案密度贴合中文自然表达，协议/路径与状态标签保留必要的英文原词（如 JSON/SSE、work_order_ops、docker 命令）。","en":"The English page shares the same brief and section sequence; copy length stays natural for English while protocol/path labels and commands keep their canonical form."},"forbiddenPatterns":["generic hero slogan over three metric cards","feature-card wall that could host any project's copy","uniform radius, shadow, and blur across every section","decorative gradients, glows, grids, or particles as ornament","browser, phone, terminal, dashboard, or editor shell frames","fake tabs, search, switches, inputs, charts, or progress controls"]},"visualReview":{"schemaVersion":"dodot-automated-design-review-v1","status":"approved","summary":"$24","checks":{"projectFit":{"passed":true,"evidence":"Both locales render the exact brief sectionSequence (hero, gate-loop-anatomy, shared-behavior-boundary, reference-workflow, when-and-when-not, run-local). Hero establishes identity, one-line value frame, GitHub CTA, and the gate-loop visual; palette (deep ink, amber gate, olive approval, paper white) and monospace/display/serif typography match the brief; desktop alternates wide-loop and narrow-plate widths. QA reports 6 sections in both locales and viewports."},"evidenceBound":{"passed":true,"evidence":"Python/MIT/foamtor/agentbridge from repository facts; 'v1.0.0 single-node stable' from summary claim; shared behaviors JSON/SSE, session lifecycle, permissions, approvals, audit, RAG ports from description; work_order_ops 4-stage loop from README/summary; caveats (production drills pending, multi-node explicit config, not real-time trading) from summary.caveats; commands 'docker compose up --build' and 'docker compose logs api' from readme.commands."},"signatureVisual":{"passed":true,"evidence":"A real radial ring SVG exists in the anatomy section (concentric circles, QUERY/RETRIEVAL/CREATE labels, an amber token, an olive approval node with a break line and stop-dot) plus a 4-step loop panel in the hero with the gate step highlighted as the interrupt. Source note states the visual is derived from work_order_ops, not a screenshot, matching the media strategy."},"mobileTransformation":{"passed":true,"evidence":"@media (max-width:820px) collapses split/when/commands grids to one column, stacks the reference workflow (wf-row) into a vertical decision-stub column with rotated arrows, single-column owned ledger, scaled radial with adjusted overlapping annotation, and no horizontal overflow (scrollWidth 390 === clientWidth 390). The hero loop is already a linear 4-step pipeline with the gate as a full-width strip, providing the 'unswirl' linear form."},"localeParity":{"passed":true,"evidence":"zh-CN and en share identical section order and heading structure (1 h1, 6 h2, 6 sections each). zh-CN keeps canonical protocol/path/command terms in English (JSON/SSE, work_order_ops, docker) and reads natural with CJK-appropriate density; en copy is natural and equivalent in meaning. Locale contract honored in both."},"antiTemplate":{"passed":true,"evidence":"No generic hero-over-three-metric-cards; the owned ledger and 4-step workflow are content-bound to project evidence, not interchangeable feature tiles. Sections alternate dark ink / paper-2 plates for rhythm rather than uniform radius/shadow. No device/browser/terminal shell frames. The visual idea is the radial gate-loop, not a gradient/grid field."},"noFakeUi":{"passed":true,"evidence":"No fake tabs, search, switches, inputs, toggles, progress bars, or gauges. Mono labels ('[ GATE ]', 'release | deny', '✓'/'—' markers) are descriptive text only; the only interactive element is the real GitHub link. QA reports fakeControlCount 0 in all locales/viewports and no device shells."}},"issues":[{"code":"mobile-unswirl-partial","severity":"warning","locale":"all","description":"The brief's mobileTransformation specifies that the polar loop 'unswirls into a linear 4-step pipeline with the gate as a full-width approval strip'. On mobile the hero already offers a linear 4-step pipeline, but the anatomy section's radial ring SVG is only scaled to fit (max-width 100%) and remains polar instead of being restructured into a linear pipeline. The transformation works but does not literally unswirl the canonical radial diagram.","repairInstruction":"In the max-width:820px media query, convert the anatomy radial diagram (or the hero loop) into a vertical 4-step linear pipeline on mobile with an un-collapsed node and the approval node rendered as a full-width gate strip, so the polar loop visibly unswirls rather than merely shrinking."},{"code":"hero-ambient-gradient","severity":"warning","locale":"all","description":"The hero uses two low-opacity radial-gradient ambient glows (0.10/0.14 alpha) behind the headline. This is very subtle and does not carry the visual idea, but it sits in the forbiddenPatterns 'decorative gradients, glows, grids, or particles as ornament' shade and could read as decorative glow.","repairInstruction":"Optionally reduce or remove the hero ambient radial-gradient backdrop, letting the radial-gate-loop panel and paper/ink depth carry the hero atmosphere, to keep the composition strictly within the brief's decoration constraints."}]},"repairEvidence":[],"provenance":{"executor":"opencode","model":"deepseek/deepseek-v4-flash","opencodeVersion":"1.1.58","promptVersion":"dodot-showcase-deepseek-v4","inputHash":"6be79cbbcc98a7e85efce38056411da58d4a65fdfda86a07841717ae7eaf4ab1","artifactHash":"238a5c60308199a0d72cd47ef608c33af88a4e1bbefbd3a4ea5c0b0d11132b73","designSkill":"dodot-showcase-design"},"qaEvidence":{"zh-CN":[{"viewport":"desktop","passed":true,"localePassed":true,"metrics":{"language":"zh-CN","h1Count":1,"h2Count":6,"sectionCount":6,"textLength":2084,"cjkCount":1059,"latinLetterCount":536,"scrollWidth":1280,"clientWidth":1280,"scrollHeight":5563,"h1Top":167.078125,"bottomInset":177.546875,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"b90723fab4b5aefed1c13752acd7684bd6de561630e83cb9b493a593e20becec","middle":"f1ccf828b1cba5f114e16a10bcfbcd3844f681cf861121e698d07d410a041ad0","bottom":"71b08034baa73e40dd5ebad50f843d8708d316a4f4f7625950c9a62d897b9ac5"},"consoleErrors":[],"networkRequests":[]},{"viewport":"mobile","passed":true,"localePassed":true,"metrics":{"language":"zh-CN","h1Count":1,"h2Count":6,"sectionCount":6,"textLength":2084,"cjkCount":1059,"latinLetterCount":536,"scrollWidth":390,"clientWidth":390,"scrollHeight":7077,"h1Top":176.640625,"bottomInset":211.40625,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"df04f1f34dd9f000eb0dab20f43a34c2d0f609f14ef089ebb62c83488a01b7d9","middle":"920cd6007e76a69a4be6c7f5b92b047e07d2b264f7e52cd62b925674aa7ad29f","bottom":"a22620fb0cb5189f21957750a44d3bdcd986a1f0b9199a37d92171128d0c4497"},"consoleErrors":[],"networkRequests":[]}],"en":[{"viewport":"desktop","passed":true,"localePassed":true,"metrics":{"language":"en","h1Count":1,"h2Count":6,"sectionCount":6,"textLength":4922,"cjkCount":0,"latinLetterCount":3893,"scrollWidth":1280,"clientWidth":1280,"scrollHeight":6159,"h1Top":186.640625,"bottomInset":178.0625,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"c097aeeca7a1e1a3192005bd65b5febc0c77c716836cfd845276d559a801d557","middle":"eb25bab718d97bf8766a0ad5fa595549c16e5695ed44c2fb2afc15f0146de5f9","bottom":"a0e5eae342f68ee6dbc46a817c8305ad688fc002315d49aab82e79a1b106eeba"},"consoleErrors":[],"networkRequests":[]},{"viewport":"mobile","passed":true,"localePassed":true,"metrics":{"language":"en","h1Count":1,"h2Count":6,"sectionCount":6,"textLength":4922,"cjkCount":0,"latinLetterCount":3893,"scrollWidth":390,"clientWidth":390,"scrollHeight":7965,"h1Top":176.640625,"bottomInset":211.5,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"556af1e947d222b96d88b6d2ce4e76530e17ac4703232857c71a23031de27685","middle":"28a8f185e7b5825ed308bf480fbeb537dcc83f0ddac57bd2281bf21dae926bf8","bottom":"c4d84688aa738700e09f4ef3ff2a7066f689ae8db2693595be6317bf0d9bff47"},"consoleErrors":[],"networkRequests":[]}]}},"generatorVersion":"opencode-1.1.58:deepseek/deepseek-v4-flash:dodot-showcase-deepseek-v4","publishedAt":"2026-08-17T15:11:31.243Z"},"rawExtraction":{"projectRevisionId":"cmt2zh05j00m96271x4yulhzx","projectVersion":1,"sources":[{"canonicalKey":"github:foamtor/agentbridge","role":"primary","revisionVersion":1,"url":"https://github.com/Foamtor/AgentBridge"}],"facts":[{"key":"repository_owner","value":"foamtor","evidence":[{"path":"repository.owner","excerpt":"foamtor"}]},{"key":"repository_name","value":"agentbridge","evidence":[{"path":"repository.name","excerpt":"agentbridge"}]},{"key":"description","value":"AgentBridge is a self-hosted, AI-readable foundation for Vibe Coding business tools. It lets an AI and your developers write the business-specific domain and tools, while the platform owns the difficult shared behavior: JSON/SSE , session lifecycle, permissions, approvals, audit, and RAG ports.","evidence":[{"path":"repository.description","excerpt":"AgentBridge is a self-hosted, AI-readable foundation for Vibe Coding business tools. It lets an AI and your developers write the business-specific domain and tools, while the platform owns the difficult shared behavior: JSON/SSE , session lifecycle, permissions, approvals, audit, and RAG ports."}]},{"key":"primary_language","value":"Python","evidence":[{"path":"repository.language","excerpt":"Python"}]},{"key":"license","value":"MIT","evidence":[{"path":"repository.license","excerpt":"MIT"}]},{"key":"source_count","value":"1","evidence":[{"path":"projectRevision.sources","excerpt":"github:foamtor/agentbridge"}]}],"taxonomy":{"category":"development_tools","subcategory":"data_ai_tools","confidence":0.65,"evidence":[{"path":"repository.description","excerpt":"AgentBridge is a self-hosted, AI-readable foundation for Vibe Coding business tools. It lets an AI and your developers write the business-specific domain and tools, while the platform owns the difficult shared behavior: JSON/SSE , session lifecycle, permissions, approvals, audit, and RAG ports."}]},"lifecycle":{"stage":"prototype","confidence":0.65,"evidence":[{"path":"readmeExcerpt","excerpt":"

# AgentBridge 🔌 **Add Agent capabilities to your business systems — knowledge retrieval, data analysis, structured output, human-in-the-loop approval — without changing a single line of existing code.** Built for [Vibe Coding](#-vibe-coding-integration): let AI coding assistants write your plugins. [简体中文](README.zh-CN.md) · [Documentation](docs/INDEX.md) · [Architecture rules](AGENTS.md) · [Vibe Coding Guide](SKILL.md) [![CI](https://github.com/Foamtor/AgentBridge/actions/w"}]},"userIntents":[{"slug":"build_software","label":"构建或改进软件","confidence":0.65,"evidence":[{"path":"repository.description","excerpt":"AgentBridge is a self-hosted, AI-readable foundation for Vibe Coding business tools. It lets an AI and your developers write the business-specific domain and tools, while the platform owns the difficult shared behavior: JSON/SSE , session lifecycle, permissions, approvals, audit, and RAG ports."}]},{"slug":"automate_work","label":"自动化工作流程","confidence":0.65,"evidence":[{"path":"repository.description","excerpt":"AgentBridge is a self-hosted, AI-readable foundation for Vibe Coding business tools. It lets an AI and your developers write the business-specific domain and tools, while the platform owns the difficult shared behavior: JSON/SSE , session lifecycle, permissions, approvals, audit, and RAG ports."}]},{"slug":"create_content","label":"创作内容或设计","confidence":0.65,"evidence":[{"path":"repository.description","excerpt":"AgentBridge is a self-hosted, AI-readable foundation for Vibe Coding business tools. It lets an AI and your developers write the business-specific domain and tools, while the platform owns the difficult shared behavior: JSON/SSE , session lifecycle, permissions, approvals, audit, and RAG ports."}]},{"slug":"analyze_data","label":"分析与理解数据","confidence":0.65,"evidence":[{"path":"repository.description","excerpt":"AgentBridge is a self-hosted, AI-readable foundation for Vibe Coding business tools. It lets an AI and your developers write the business-specific domain and tools, while the platform owns the difficult shared behavior: JSON/SSE , session lifecycle, permissions, approvals, audit, and RAG ports."}]},{"slug":"learn_explore","label":"学习与探索","confidence":0.65,"evidence":[{"path":"repository.description","excerpt":"AgentBridge is a self-hosted, AI-readable foundation for Vibe Coding business tools. It lets an AI and your developers write the business-specific domain and tools, while the platform owns the difficult shared behavior: JSON/SSE , session lifecycle, permissions, approvals, audit, and RAG ports."}]}],"readme":{"headings":["AgentBridge","🏗️ How it works without changing your code","🔩 Under the hood","🤖 Vibe Coding Integration","🎯 When to use it","⚠️ When NOT to use it","🚀 Run it and see for yourself","📋 A complete example","📦 What the platform does","🗂️ Repository structure","📊 Current status","🤝 Contributing","📄 License"],"commands":["docker compose up --build","docker compose logs api"]},"changeSet":[{"field":"repository","beforeHash":"258c0b1208f6e6ea63f2deb9c75621855a244aedb408ae6ffcba2727c2210184","afterHash":"331d77b48a56915218a94bc3170a3e63295889bafde773f321b28312f6b89c29"}]},"publishedAt":"2026-08-16T12:38:44.949Z","updatedAt":"2026-08-21T13:26:50.138Z","summary":{"id":"cmt2zh05n00mb6271lm6cdq4p","projectType":"product","oneLine":"自托管的 Vibe Coding 业务工具基础平台","overview":"AgentBridge 是一个自托管的、AI 可读的 Vibe Coding 业务工具基础平台。它允许 AI 和开发者编写业务领域和工具，而平台负责共享的复杂行为，如 JSON/SSE、会话生命周期、权限、审批、审计和 RAG 端口。无需修改现有代码即可为业务系统添加 Agent 能力，包括知识检索、数据分析、结构化输出和人工审批。","highlights":["无需修改现有代码","支持人工审批","提供 Vibe Coding 指南","包含参考实现"],"bestFor":["已有 API 或数据库的系统","希望快速验证 AI 能力的团队","需要共享治理的多系统环境"],"caveats":["生产部署演练等后续工作未完成","多实例扩展需要显式配置","实时交易场景不适用"],"categorySuggestion":"development_tools","subcategorySuggestion":"mcp_integrations","lifecycleStages":["backend_data","deployment_growth"],"audienceIntents":["build","integrate","evaluate"],"confidence":0.8,"evidence":[{"claim":"AgentBridge 是一个自托管的、AI 可读的 Vibe Coding 业务工具基础平台，允许 AI 和开发者编写业务领域和工具，而平台负责共享的复杂行为，如 JSON/SSE、会话生命周期、权限、审批、审计和 RAG 端口。","sourceType":"description","locator":"https://github.com/Foamtor/AgentBridge"},{"claim":"README 声称无需修改现有代码即可为业务系统添加 Agent 能力，包括知识检索、数据分析、结构化输出和人工审批。","sourceType":"readme","locator":"https://github.com/Foamtor/AgentBridge"},{"claim":"项目包含一个参考实现 work_order_ops，演示了完整的业务循环，包括查询、知识检索、创建工单和审批。","sourceType":"readme","locator":"https://github.com/Foamtor/AgentBridge"},{"claim":"项目提供 Vibe Coding 集成指南 SKILL.md，指导 AI 助手编写插件。","sourceType":"readme","locator":"https://github.com/Foamtor/AgentBridge"},{"claim":"项目状态为 v1.0.0 单节点稳定版，但生产部署演练、迁移/恢复、具体 IdP 集成、多实例验证和包发布是后续工作。","sourceType":"readme","locator":"https://github.com/Foamtor/AgentBridge"}],"publishedAt":"2026-08-17T13:19:58.641Z"},"github":{"url":"https://github.com/Foamtor/AgentBridge","ownerAvatarUrl":"https://avatars.githubusercontent.com/u/11826083?v=4","starsCount":3,"forksCount":0,"language":"Python","syncedAt":"2026-08-17T13:18:24.047Z","license":"MIT","defaultBranch":"main","archived":false,"disabled":false,"lastChangedAt":"2026-08-14T11:41:48.000Z","missing":false},"tags":[],"updates":[]},"entryPoint":"discovery","claim":{"state":"unclaimed","verified":false},"initialView":"beautified","community":{"viewer":null,"state":{"likesCount":0,"creativeCount":0,"followersCount":0,"commentsCount":0,"liked":false,"creative":false,"followed":false},"comments":{"comments":[],"nextCursor":null}}}]}],"$L25"]