Skip to content
Back to search

agentfield

Build, run and scale AI agents like API and microservices

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today

Check it deeper on your machine

It has changed since its scan on 11 Nov 2026.

As a Claude Code plugin, install it and then ask Claude Code:

/plugin marketplace add keithmackay/mackayi/plugin install scanskillsafety@mackayiDo a deep scan of https://github.com/Agent-Field/agentfield with scanskillsafety

Or from a terminal, clone it and run it:

git clone https://github.com/keithmackay/scanskillsafety-skill ~/.claude/skills/scanskillsafetynode ~/.claude/skills/scanskillsafety/cli.cjs --deep 'https://github.com/Agent-Field/agentfield'

It checks the exact commit: every file, its packages against OSV (api.osv.dev), how it compares with this page's scan (findsafeskills.com receives the repository's address), and, if you agree, Claude reading every file with your own Claude subscription. It never runs the code.

Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/agent-field--agentfield/badge.svg)](/listings/agent-field--agentfield)

HTML

<a href="/listings/agent-field--agentfield"><img src="/api/listings/agent-field--agentfield/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 658 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

This file has changed since it was scanned, so this is not exactly the text that was scanned, and no line is marked. The findings are listed with the safety scan above.

  1. <div align="center">
  2. # AgentField — The AI Backend
  3. ### **Build agents like APIs. Run ten thousand of them like microservices.**
  4. *One request fans out to thousands of agents. The control plane queues, retries, and traces every branch.*
  5. [![Stars](https://img.shields.io/github/stars/Agent-Field/agentfield?style=flat&logo=github&logoColor=e8e5dc&color=0c0b09&labelColor=8b7355)](https://github.com/Agent-Field/agentfield/stargazers)
  6. [![License](https://img.shields.io/badge/license-Apache%202.0-0c0b09.svg?style=flat&labelColor=8b7355)](LICENSE)
  7. [![Downloads](https://img.shields.io/endpoint?url=https%3A%2F%2Fgist.githubusercontent.com%2Fsantoshkumarradha%2Fd98e2ad73502b4075f6a5f0ae4f5cae5%2Fraw%2Fbadge.json&style=flat&logo=download&logoColor=e8e5dc&color=0c0b09&labelColor=8b7355&cacheSeconds=3600)](https://github.com/Agent-Field/agentfield)
  8. [![Coverage](https://img.shields.io/endpoint?url=https%3A%2F%2Fgist.githubusercontent.com%2Fsantoshkumarradha%2F433fb09c2d54c3c2589125cfd3eb14a2%2Fraw%2Fbadge.json&style=flat&logo=github&logoColor=e8e5dc&color=0c0b09&labelColor=8b7355&cacheSeconds=3600)](https://github.com/Agent-Field/agentfield/actions/workflows/coverage.yml)
  9. [![Last Commit](https://img.shields.io/github/last-commit/Agent-Field/agentfield?style=flat&logo=git&logoColor=e8e5dc&color=0c0b09&labelColor=8b7355)](https://github.com/Agent-Field/agentfield/commits/main)
  10. [![Discord](https://img.shields.io/badge/discord-join%20us-0c0b09.svg?style=flat&labelColor=8b7355&logo=discord&logoColor=e8e5dc)](https://discord.gg/aBHaXMkpqh)
  11. [![Featured in Awesome AI Agents](https://img.shields.io/badge/Featured_in-Awesome_AI_Agents-2ea44f)](https://github.com/NipunaRanasinghe/awesome-ai-agents)
  12. <!-- All outbound agentfield.ai links must use UTM parameters, not direct URLs. See assets/utm-links.csv for the full list — target URLs there are for verification and reference only, always use the UTM version in the README. Update both the README and the CSV when adding or changing links. -->
  13. **[Docs](https://agentfield.ai/docs/learn?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-docs)** · **[Quick Start](https://agentfield.ai/docs/learn/quickstart?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-quickstart)** · **[Python SDK](https://agentfield.ai/docs/reference/sdks/python?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-python-sdk)** · **[Go SDK](https://agentfield.ai/docs/reference/sdks/go?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-go-sdk)** · **[TypeScript SDK](https://agentfield.ai/docs/reference/sdks/typescript?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-typescript-sdk)** · **[REST API](https://agentfield.ai/docs/reference/sdks/rest-api?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-rest-api)** · **[Examples](#built-with-agentfield)** · **[Discord](https://discord.gg/aBHaXMkpqh)**
  14. </div>
  15. <div align="center">
  16. <a href="https://agentfield.ai/docs/build/intelligence/harness?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-harness-banner">
  17. <img src="assets/harness-banner.png" alt="Now includes Harness Orchestration — multi-turn coding agents with CodeAF, Claude Code, Codex, Gemini CLI, and OpenCode" width="100%" />
  18. </a>
  19. </div>
  20. AgentField is an open-source control plane that lets you build AI agents callable by any service in your stack - frontends, backends, other agents, cron jobs - just like any other API. You write agent logic in Python, Go, or TypeScript. AgentField turns it into production infrastructure: routing, coordination, memory, async execution, and observability. Every function becomes a REST endpoint, and the same code scales from one agent on your laptop to ten thousand in a single workflow: the control plane handles the fan-out, the queues, and the retries.
  21. <div align="center">
  22. https://github.com/user-attachments/assets/9fb7b1cf-26de-4b9b-9ba2-917252cc26ec
  23. <sub><b>One prompt → a running containerized production ready multi-agent backend.</b> No glue code, start using the agent API!</sub>
  24. </div>
  25. ## Build production agents with a prompt.
  26. **Describe the system in one line. Get a production-ready multi-agent backend.** Works in Claude Code, Codex, Gemini CLI, OpenCode, Aider, Windsurf, and Cursor.
  27. ```bash
  28. curl -fsSL https://agentfield.ai/install.sh | bash
  29. ```
  30. The installer also installs the `codeaf` coding harness in
  31. `$AGENTFIELD_HOME/bin` (`~/.agentfield/bin` by default), so harness-backed agents
  32. work out of the box; skip it with `--no-codeaf` (`--no-aforge` remains an alias). Staging uses
  33. `~/.agentfield-staging`; an explicit `AGENTFIELD_HOME` selects the production
  34. managed home. `AGENTFIELD_INSTALL_DIR` selects the `af` directory; CodeAF
  35. always uses the managed home's `bin`. Repair it with `af codeaf ensure`.
  36. The provider name `aforge` remains an alias; set `AFORGE_BIN` and select
  37. `provider="aforge"` to pin the legacy adapter. CodeAF state stays isolated in
  38. `$AGENTFIELD_HOME/codeaf`. See [harness providers](docs/harness-providers.md)
  39. for binary overrides, mirror/skip switches, and telemetry controls.
  40. On macOS the installer also registers the control plane to start at login (under
  41. launchd) and adds a menu-bar icon. Stop it with `af service stop` or the menu-bar
  42. icon — a plain `kill` looks like a crash and it restarts. `af service status`
  43. shows health and in-flight work; install with `--no-tray` to skip this entirely.
  44. Then in your coding agent, paste any spec with /agentfield :
  45. ```text
  46. /agentfield Build a claims processor with risk scoring, pattern detection,
  47. and human approval for low-confidence decisions.
  48. ```
  49. You get a Docker Compose stack wired up end-to-end — the agent, the control plane, and a production ready REST API endpoint you can paste and `curl` into a terminal to try it. [See it in action →](https://agentfield.ai/docs/learn/build-with-claude-code?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-prompt-to-production)
  50. ## The DX you get
  51. *Plain Python (or Go / TypeScript) functions. No DSL, no YAML, no graph wiring.*
  52. ```python
  53. import asyncio
  54. from agentfield import Agent, AIConfig
  55. from pydantic import BaseModel
  56. app = Agent(
  57. node_id="researcher",
  58. version="1.0.0",# Canary deploys, A/B testing, blue-green rollouts
  59. ai_config=AIConfig(model="anthropic/claude-sonnet-4-20250514"),
  60. )
  61. class SubQuestions(BaseModel):
  62. questions: list[str]
  63. @app.reasoner(tags=["research"])
  64. async def research(question: str, depth: int = 0, model: str | None = None) -> dict:
  65. if depth >= 3: # depth cap keeps fan-out bounded
  66. answer = await app.ai(system="Answer directly and concisely.", user=question, model=model)
  67. return {"question": question, "answer": answer}
  68. # Break the question into sub-questions
  69. plan = await app.ai(
  70. system="Break this into 3-5 independent sub-questions.",
  71. user=question, schema=SubQuestions, model=model,
  72. )
  73. # Fan out: each sub-question recurses on this same agent, through the control plane
  74. branches = await asyncio.gather(*[
  75. app.call(f"{app.node_id}.research", question=q, depth=depth + 1, model=model)
  76. for q in plan.questions
  77. ])
  78. # Synthesize the branches back into one answer
  79. synthesis = await app.ai(system="Synthesize these findings.", user=str(branches), model=model)
  80. return {"question": question, "answer": synthesis, "branches": branches}
  81. app.run()
  82. # This single line exposes: POST /api/v1/execute/researcher.research
  83. # One request fans out to thousands of agents. The control plane queues, retries, and traces
  84. # every branch. No broker, no queue setup, no timeout.
  85. ```
  86. > **What you just saw:** `app.ai()` calls an LLM and returns structured output. `app.call()` routes to other agents (or back to itself) through the control plane, so recursion becomes distributed fan-out. `asyncio.gather()` runs every branch in parallel. `app.run()` auto-exposes everything as REST. [Read the full docs →](https://agentfield.ai/docs/learn?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-read-full-docs)
  87. <details>
  88. <summary><b>Need approvals, audit trails, and governance? (the enterprise sample)</b></summary>
  89. ```python
  90. from agentfield import Agent, AIConfig
  91. from pydantic import BaseModel
  92. app = Agent(
  93. node_id="claims-processor",
  94. version="2.1.0",# Canary deploys, A/B testing, blue-green rollouts
  95. ai_config=AIConfig(model="anthropic/claude-sonnet-4-20250514"),
  96. )
  97. class Decision(BaseModel):
  98. action: str# "approve", "deny", "escalate"
  99. confidence: float
  100. reasoning: str
  101. @app.reasoner(tags=["insurance", "critical"])
  102. async def evaluate_claim(claim: dict) -> dict:
  103. # Structured AI judgment - returns typed Pydantic output
  104. decision = await app.ai(
  105. system="Insurance claims adjuster. Evaluate and decide.",
  106. user=f"Claim #{claim['id']}: {claim['description']}",
  107. schema=Decision,
  108. )
  109. if decision.confidence < 0.85:
  110. # Human approval - suspends execution, notifies via webhook, resumes when approved
  111. await app.pause(
  112. approval_request_id=f"claim-{claim['id']}",
  113. approval_request_url=f"https://internal.acme.com/approvals/claim-{claim['id']}",
  114. expires_in_hours=48,
  115. )
  116. # Route to the next agent - traced through the control plane
  117. await app.call("notifier.send_decision", input={
  118. "claim_id": claim["id"],
  119. "decision": decision.model_dump(),
  120. })
  121. return decision.model_dump()
  122. app.run()
  123. # This single line exposes: POST /api/v1/execute/claims-processor.evaluate_claim
  124. # The agent auto-registers with the control plane, gets a cryptographic identity, and every
  125. # execution produces a verifiable, tamper-proof audit trail.
  126. ```
  127. > **What you just saw:** `app.ai()` calls an LLM and returns structured output. `app.pause()` suspends for [human approval](https://agentfield.ai/docs/build/execution/human-in-the-loop?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-human-in-the-loop). `app.call()` routes to other agents through the control plane. `app.run()` auto-exposes everything as REST. [Read the full docs →](https://agentfield.ai/docs/learn?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-read-full-docs)
  128. </details>
  129. <details>
  130. <summary><b>Prefer to scaffold by hand? (Python / Go / TypeScript / Docker)</b></summary>
  131. ```bash
  132. af init my-agent --defaults # Scaffold agent
  133. cd my-agent && pip install -r requirements.txt
  134. af server # Terminal 1 → Dashboard at http://localhost:8080
  135. python main.py # Terminal 2 → Agent auto-registers
  136. ```
  137. ```bash
  138. # Call your agent
  139. curl -X POST http://localhost:8080/api/v1/execute/my-agent.demo_echo \
  140. -H "Content-Type: application/json" \
  141. -d '{"input": {"message": "Hello!"}}'
  142. ```
  143. ```bash
  144. # Go
  145. af init my-agent --defaults --language go && cd my-agent && go run .
  146. # TypeScript
  147. af init my-agent --defaults --language typescript && cd my-agent && npm install && npm run dev
  148. # Docker (control plane only)
  149. docker run -p 8080:8080 agentfield/control-plane:latest
  150. ```
  151. [Deployment guide →](https://agentfield.ai/docs/reference/deploy?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-deploy) for Docker Compose and production setups, or use the repository's [Kubernetes deployment guide](docs/deploying-on-kubernetes.md).
  152. </details>
  153. ## See it in action
  154. <!-- TODO: replace static UI.png with a short gif of a live fan-out run in the DAG view -->
  155. <div align="center">
  156. <img src="assets/UI.png" alt="AgentField Dashboard" width="100%" />
  157. <br/>
  158. <sub>Real-time workflow DAGs · Execution traces · Agent fleet management · Audit trails</sub>
  159. </div>
  160. ## How AgentField fits in your stack
  161. Most agent tools help you **write** agent logic. AgentField is what **runs** it in production: the layer that makes agents callable by other software, durable across failures, and observable when one request fans out to a thousand branches. Keep the framework you already use for authoring; a reasoner is a plain function, so existing LangGraph or CrewAI code can run inside one.
  162. | | **Frameworks**<br><sub>LangChain · CrewAI · PydanticAI · OpenAI Agents SDK</sub> | **Workflow engines**<br><sub>Temporal · Airflow</sub> | **Visual builders**<br><sub>n8n · Zapier</sub> | **AgentField** |
  163. |---|:-:|:-:|:-:|:-:|
  164. | Build agent logic (prompts, tools, structured output) | ● | — | — | ● |
  165. | Prebuilt chains, retrievers, integrations | ● | — | ◐ | [◐](https://agentfield.ai/docs/integrations?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-integrations) |
  166. | Production REST APIs out of the box | — | ◐ | ● | ● |
  167. | Async + retries + webhooks | — | ● | ◐ | **●** |
  168. | Memory scopes (global · actor · session · workflow) | ◐ | — | — | ● |
  169. | Service discovery + cross-agent calls | — | — | — | **●** |
  170. | Distributed agents (register from anywhere, one mesh) | — | ◐ | — | **●** |
  171. | Coding agents as functions (Claude Code · Codex · CLI) | — | — | — | **●** |
  172. | Agent identity, access policies, signed audit trails | — | — | — | ● |
  173. | Fleet observability (DAGs · metrics · traces) | — | ◐ | — | ● |
  174. | Multi-language SDKs (Python · Go · TypeScript) | ◐ | ● | — | ● |
  175. <sub>● full · ◐ partial · — not the focus</sub>
  176. **Prototype in whatever you like. The moment a second service needs to call your agent, put it on AgentField.** That is the point where you would otherwise start writing queues, retries, discovery, and tracing by hand.
  177. [Full comparison & decision guide →](https://agentfield.ai/docs/learn/vs-frameworks?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-vs-frameworks)
  178. ## How it scales
  179. The control plane is a stateless Go service. You put more of them behind a load balancer and the fleet grows horizontally. Work is admitted into a bounded in-process queue with backpressure (`429`/`503` plus `Retry-After`). On graceful shutdown, in-flight executions are terminated with `status_reason` `control_plane_shutdown` rather than silently dropped.
  180. | Property | What it means |
  181. |---|---|
  182. | Stateless Go control plane | Horizontal scaling behind a load balancer. Add replicas to add capacity. |
  183. | Bounded in-process admission | Backpressure returns `429`/`503` with `Retry-After`; graceful shutdown records `control_plane_shutdown`. |
  184. | Async execution | Webhooks and SSE, no timeout limits. A single run can go for hours or days. |
  185. | Backpressure | Queue-depth limits and circuit breakers keep a fan-out from overwhelming downstream agents. |
  186. | Routing overhead | Roughly 100-200ms per cross-agent hop. It matters when a branch does little work per hop, so keep hops coarse when latency is tight. |
  187. Two examples already run at this load. The [deep-research engine](https://agentfield.ai/github/deepresearch/?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-deepresearch-repo) fanned out 10,000+ agent invocations in one workflow. The [security auditor](https://agentfield.ai/github/sec-af/?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-sec-af-repo) runs 250 coordinated agents per audit.
  188. [Deployment guide →](https://agentfield.ai/docs/reference/deploy?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-deploy-scale) for Docker Compose, Kubernetes, and production setups.
  189. ## What You Get
  190. **Build** - Python, Go, or TypeScript. Every function becomes a REST endpoint.
  191. - **[Reasoners & Skills](https://agentfield.ai/docs/build/building-blocks/reasoners?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-reasoners)** - `@app.reasoner()` for AI judgment, `@app.skill()` for deterministic code
  192. - **[Structured AI](https://agentfield.ai/docs/reference/sdks/python?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-structured-ai)** - `app.ai(schema=MyModel)` → typed Pydantic/Zod output from any LLM
  193. - **[Harness](https://agentfield.ai/docs/build/intelligence/harness?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-harness)** - `app.harness("Fix the bug")` dispatches multi-turn tasks to CodeAF, AgentField's own coding harness — no setup. Add `provider="claude-code"` (or `codex`, `gemini`, `opencode`, `pi`, `omp`) to orchestrate someone else's.
  194. - **[Cross-Agent Calls](https://agentfield.ai/docs/build/coordination/cross-agent-calls?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-cross-agent-calls)** - `app.call("other-agent.func")` routes through the control plane with full tracing
  195. - **[Discovery](https://agentfield.ai/docs/reference/sdks/python?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-discovery)** - `app.discover(tags=["ml*"])` finds agents and capabilities across the mesh. `tools="discover"` lets LLMs auto-invoke them.
  196. - **[Memory](https://agentfield.ai/docs/build/coordination/shared-memory?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-memory)** - `app.memory.set()` / `.get()` / `.similarity_search()` - KV + vector search, four scopes, no Redis needed
  197. **Scale** - Production infrastructure for non-deterministic AI.
  198. - **[Async Execution](https://agentfield.ai/docs/build/execution/async?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-async-execution)** - Fire-and-forget with webhooks, SSE streaming, retries. No timeout limits - agents run for hours or days.
  199. - **[Canary Deployments](https://agentfield.ai/docs/learn/features?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-canary-deployments)** - Traffic weight routing, A/B testing, blue-green deploys. Roll out agent versions at 5% → 50% → 100%.
  200. - **[Human-in-the-Loop](https://agentfield.ai/docs/build/execution/human-in-the-loop?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-human-in-the-loop)** - `app.pause()` suspends execution for human approval. Crash-safe, durable, audited.
  201. - **[Observability](https://agentfield.ai/docs/learn/features?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-observability)** - Automatic workflow DAGs, Prometheus `/metrics`, structured logs, execution timeline.
  202. **Govern** - IAM for AI agents. Every agent gets a cryptographic identity. Identity, access control, and audit trails - built in.
  203. - **[Cryptographic Identity](https://agentfield.ai/docs/build/governance/identity?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-crypto-identity)** - Every agent gets a W3C DID (decentralized identifier) - not a shared API key. Agents authenticate to each other the way services authenticate with mTLS, but with cryptographic signatures that travel with the agent.
  204. - **[Verifiable Credentials](https://agentfield.ai/docs/build/governance/credentials?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-verifiable-credentials)** - Tamper-proof receipt for every execution. Offline-verifiable: `af vc verify audit.json`.
  205. - **[Policy Enforcement](https://agentfield.ai/docs/build/governance/policy?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-policy-enforcement)** - Tag-based policy gates with cryptographic verification. "Only agents tagged 'finance' can call this" - enforced by infrastructure, not prompts.
  206. [See the full production-ready feature set →](https://agentfield.ai/docs/learn/features?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-full-features)
  207. <div align="center">
  208. <img src="assets/features-strip.png" alt="90+ Production Features" width="100%" />
  209. </div>
  210. <details>
  211. <summary><h4 align="center">▼ Click to expand full capabilities</h4></summary>
  212. #### AI & LLM
  213. | Feature | How |
  214. |---|---|
  215. | Structured output (Pydantic/Zod) | `app.ai(schema=MyModel)` |
  216. | Multi-turn coding agents | `app.harness("task")` — CodeAF by default |
  217. | Orchestrate another harness | `app.harness("task", provider="claude-code")` (also `codex`, `gemini`, `opencode`, `pi`, `omp`) |
  218. | LLM auto-discovers agents and tools | `app.ai(tools="discover")` |
  219. | Multimodal (text, image, audio) | `app.ai("Describe", image_url="...")` |
  220. | Streaming responses | `app.ai("...", stream=True)` |
  221. | 100+ LLMs via LiteLLM | `AIConfig(model="anthropic/claude-sonnet-4-20250514")` |
  222. | Temperature, max tokens, format | `app.ai(..., temperature=0.2)` |
  223. #### Agent Mesh & Discovery
  224. | Feature | How |
  225. |---|---|
  226. | Cross-agent calls with tracing | `app.call("agent.func", input={...})` |
  227. | Discover agents by tag (wildcards) | `app.discover(tags=["ml*"])` |
  228. | Discover by health status | `app.discover(health_status="active")` |
  229. | Agent routers (namespacing) | `AgentRouter(prefix="billing")` |
  230. | Auto context propagation | Workflow, session, actor IDs forwarded |
  231. | Parallel agent execution | `asyncio.gather(app.call(...), ...)` |
  232. | Auto-registration on startup | Service mesh with zero config |
  233. #### Execution Engine
  234. | Feature | How |
  235. |---|---|
  236. | Sync execution (REST) | [`POST /api/v1/execute/{agent}.{func}`](docs/api/EXECUTE.md) |
  237. | Async (fire-and-forget) | [`POST /api/v1/execute/async/{agent}.{func}`](docs/api/EXECUTE.md) |
  238. | Webhooks + HMAC-SHA256 signing | `AsyncConfig(webhook_url="...", secret="...")` |
  239. | SSE streaming (real-time) | `/api/v1/execute/stream/{id}` |
  240. | No timeout limits (hours/days) | Control plane allows unlimited duration |
  241. | Execution polling | `GET /api/v1/executions/{id}` |
  242. | Restart/replay | [`POST /api/v1/executions/{id}/restart`](docs/api/EXECUTION_RESTART.md) |
  243. | Run naming, labels, links | [`POST /api/v1/runs/{id}/metadata`](docs/api/RUN_METADATA.md) |
  244. | Batch status checks | `POST /api/v1/executions/batch-status` |
  245. | Progress updates mid-execution | Intermediate payloads during long tasks |
  246. | Auto retries + exponential backoff | Transparent - control plane handles |
  247. | Backpressure + queue depth limits | Fair scheduling, circuit breakers |
  248. | Bounded in-process queue | Backpressure and explicit graceful-shutdown termination |
  249. #### Memory (Distributed State)
  250. | Feature | How |
  251. |---|---|
  252. | Key-value storage | `app.memory.set(key, value)` / `.get(key)` |
  253. | Vector search (semantic) | `app.memory.similarity_search(embedding, top_k=5)` |
  254. | Four scopes | Global, actor, session, workflow |
  255. | Reactive memory events | `@app.memory.on_change("order_*")` |
  256. | Metadata filtering | Filter stored values by metadata |
  257. | Zero dependencies | Built into control plane - no Redis |
  258. #### Human-in-the-Loop
  259. | Feature | How |
  260. |---|---|
  261. | Durable pause/resume | `await app.pause(reason="...")` |
  262. | Approval workflows with UI | `approval_request_url` for reviewers |
  263. | Configurable timeouts | `expires_in_hours=24` + auto-escalation |
  264. | Crash-safe state | Survives agent restarts |
  265. #### Canary Deployments & Versioning
  266. | Feature | How |
  267. |---|---|
  268. | Traffic weight routing | 5% → 50% → 100% rollouts |
  269. | A/B testing | 50/50 splits with `X-Routed-Version` |
  270. | Blue-green deployments | Instant weight switch, zero downtime |
  271. | Per-version health tracking | Unhealthy versions auto-removed |
  272. | Agent lifecycle states | pending → starting → ready → degraded → offline |
  273. #### Identity & Governance
  274. | Feature | How |
  275. |---|---|
  276. | Cryptographic identity per agent | Auto-generated W3C DID + Ed25519 keys |
  277. | Verifiable Credentials | Tamper-proof receipt per execution |
  278. | Offline VC verification | `af vc verify audit.json` |
  279. | Tag-based access policies | ALLOW/DENY rules on caller → target tags |
  280. | Cryptographically signed requests | Ed25519 signatures on cross-agent calls |
  281. | VC hierarchy (3 tiers) | Platform → Node → Function control |
  282. | Agent notes (audit log) | `app.note("Decision", tags=["critical"])` |
  283. | Non-repudiation | Cryptographic proof of actions |
  284. | Permission request workflows | Auto-created when access denied |
  285. #### Observability & Fleet Management
  286. | Feature | How |
  287. |---|---|
  288. | Automatic DAG visualization | Workflow graphs in dashboard |
  289. | Prometheus metrics | `/metrics` out of the box |
  290. | Structured JSON logging | Automatic from SDK |
  291. | Execution timeline | Chronological decision trace |
  292. | Health checks (K8s-ready) | `/health`, `/ready` endpoints |
  293. | Correlation IDs | `X-Workflow-ID`, `X-Execution-ID` |
  294. | Workflow DAG API | `GET /api/v1/workflows/{id}/dag` |
  295. | Agent heartbeat monitoring | Auto health status transitions |
  296. #### Harness (Multi-turn Coding Agents)
  297. | Feature | How |
  298. |---|---|
  299. | Zero-setup default harness | CodeAF (`codeaf`), installed alongside `af` |
  300. | Swap the worker, keep the loop | `provider="claude-code"` \| `"codex"` \| `"gemini"` \| `"opencode"` \| `"pi"` \| `"omp"` |
  301. | Fleet-wide default override | `AGENTFIELD_HARNESS_PROVIDER=codex` |
  302. | Schema-constrained output | `schema=ResultModel` (Pydantic/Zod) |
  303. | Cost capping | `max_budget_usd=3.0` |
  304. | Turn limiting | `max_turns=100` |
  305. | Tool access control | `tools=["Read", "Write", "Bash"]` |
  306. | Environment injection | `env={"KEY": "value"}` |
  307. | System prompt override | `system_prompt="..."` |
  308. | OpenCode per-run configuration | All SDKs preserve caller `OPENCODE_CONFIG_CONTENT`; `AGENTFIELD_OPENCODE_STEPS` overrides the default `500` steps |
  309. | OpenCode prompt compatibility | All SDKs support `AGENTFIELD_OPENCODE_INLINE_SYSTEM_PROMPT=1` for the opt-in inline rollback |
  310. | Provider-agnostic reasoning variants | `variant="high"` or a `#high` model suffix |
  311. | Multi-layer output recovery | Cosmetic repair → retry → full retry |
  312. #### Connector API (Fleet Management)
  313. | Feature | How |
  314. |---|---|
  315. | Remote agent management | `/connector/reasoners` |
  316. | Version traffic control | `/connector/.../weight` |
  317. | Bearer token auth | `AGENTFIELD_CONNECTOR_TOKEN` |
  318. | Air-gapped deployment | Outbound WebSocket only |
  319. #### Developer Experience
  320. | Feature | How |
  321. |---|---|
  322. | CLI scaffolding | `af init my-agent --defaults --language python\|go\|typescript` |
  323. | Local dev with dashboard | `af server` → http://localhost:8080 |
  324. | Hot reload | `af dev` auto-detects changes |
  325. | Auto-REST from decorators | Every `@app.reasoner()` → `POST /api/v1/execute/...` |
  326. | Python, Go, TypeScript SDKs | Native patterns per language |
  327. | MCP server integration | `af add --mcp --url <server>` |
  328. | Zero-setup MCP for AI harnesses | Control plane serves MCP at `<server>/mcp` → [docs](docs/mcp-integration.md) |
  329. | Config storage API | `POST /api/v1/configs/:key` - database-backed |
  330. | Docker + Kubernetes ready | Stateless control plane, horizontal scaling |
  331. [Explore all features in detail →](https://agentfield.ai/docs/learn/features?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-explore-features)
  332. </details>
  333. ## Built With AgentField
  334. Each of these is a real, installable agent node. With a control plane running, drop any of them into your setup with a single command — `af install <repo-url>` clones the repo, isolates its dependencies, and registers the node. You're prompted once for shared secrets like `OPENROUTER_API_KEY` (stored encrypted and reused across every node), then the node's reasoners are callable over REST or with `af call`:
  335. ```bash
  336. af install https://github.com/Agent-Field/SWE-AF # autonomous engineering team → node: swe-planner
  337. af install https://github.com/Agent-Field/sec-af # security auditor → node: sec-af
  338. af install https://github.com/Agent-Field/cloudsecurity-af # cloud / IaC security scanner → node: cloudsecurity
  339. af install https://github.com/Agent-Field/pr-af # agentic code review → node: pr-af
  340. af install https://github.com/Agent-Field/person-to-brief-af # cited person intelligence → node: person-to-brief-af
  341. af run swe-planner # start a node (prompts once for required secrets)
  342. af call swe-planner.build --in '{"goal": "Add JWT auth", "repo_url": "https://github.com/user/my-repo"}'
  343. ```
  344. Full walkthrough — authoring, installing, and configuring nodes: [Installing agent nodes →](docs/installing-agent-nodes.md).
  345. <table>
  346. <tr>
  347. <td align="center" width="50%">
  348. <a href="https://agentfield.ai/github/swe-af/?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-swe-af-repo">
  349. <img src="assets/examples/autonomous-engineering-team.png" alt="Autonomous Engineering Team" />
  350. </a>
  351. <br/>
  352. <b>Autonomous Engineering Team</b>
  353. <br/>
  354. <sub>One API call spins up PM, architect, coders, QA, reviewers - hundreds of coordinated agents that plan, build, test, and ship.</sub>
  355. <br/><br/>
  356. <a href="https://agentfield.ai/github/swe-af/?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-swe-af-repo">View project →</a>
  357. </td>
  358. <td align="center" width="50%">
  359. <a href="https://agentfield.ai/github/deepresearch/?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-deepresearch-repo">
  360. <img src="assets/examples/deep-research-engine.png" alt="Deep Research Engine" />
  361. </a>
  362. <br/>
  363. <b>Deep Research Engine</b>
  364. <br/>
  365. <sub>Recursive research backend. Spawns parallel agents, evaluates quality, generates deeper agents, and recurses -10,000+ agents per query.</sub>
  366. <br/><br/>
  367. <a href="https://agentfield.ai/github/deepresearch/?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-deepresearch-repo">View project →</a>
  368. </td>
  369. </tr>
  370. <tr>
  371. <td align="center" width="50%">
  372. <a href="https://agentfield.ai/github/mongodb/?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-mongodb-repo">
  373. <img src="assets/examples/reactive-database-layer.png" alt="Reactive MongoDB Intelligence" />
  374. </a>
  375. <br/>
  376. <b>Reactive MongoDB Intelligence</b>
  377. <br/>
  378. <sub>Atlas Triggers + agent reasoning. Documents arrive raw and leave enriched - risk scores, pattern detection, evidence chains.</sub>
  379. <br/><br/>
  380. <a href="https://agentfield.ai/github/mongodb/?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-mongodb-repo">View project →</a>
  381. </td>
  382. <td align="center" width="50%">
  383. <a href="https://agentfield.ai/github/sec-af/?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-sec-af-repo">
  384. <img src="assets/examples/ai-security-auditor.png" alt="Autonomous Security Audit" />
  385. </a>
  386. <br/>
  387. <b>Autonomous Security Audit</b>
  388. <br/>
  389. <sub>250 coordinated agents trace every vulnerability source-to-sink and adversarially verify each finding. Confirmed exploits, not pattern flags.</sub>
  390. <br/><br/>
  391. <a href="https://agentfield.ai/github/sec-af/?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-sec-af-repo">View project →</a>
  392. </td>
  393. </tr>
  394. <tr>
  395. <td align="center" width="50%">
  396. <a href="https://agentfield.ai/github/cloudsecurity/?utm_source=github-readme&utm_campaign=github-readme&utm_content=cloudsec&utm_id=github-readme-cloudsec-repo">
  397. <img src="assets/examples/cloud-security.png" alt="CloudSecurity AF" />
  398. </a>
  399. <br/>
  400. <b>CloudSecurity AF</b>
  401. <br/>
  402. <sub>AI-native cloud infrastructure security scanner that performs shift-left attack path analysis directly from IaC, prioritizing the most dangerous risk chains before deployment.</sub>
  403. <br/><br/>
  404. <a href="https://agentfield.ai/github/cloudsecurity/?utm_source=github-readme&utm_campaign=github-readme&utm_content=cloudsec&utm_id=github-readme-cloudsec-repo">View project →</a>
  405. </td>
  406. <td align="center" width="50%">
  407. <a href="https://agentfield.ai/github/pr-af/?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-pr-af-repo">
  408. <img src="assets/examples/agentic-pr-reviewer-hero.png" alt="Agentic PR Reviewer — #1 open-source on Code-Review-Bench" />
  409. </a>
  410. <br/>
  411. <b>Agentic PR Reviewer</b>
  412. <br/>
  413. <sub>#1 open-source reviewer on Code-Review-Bench - 0.706 golden recall across 42 tools compared, at ~10x lower cost per review. Builds a custom review strategy for every PR, then adversarially challenges its own findings.</sub>
  414. <br/><br/>
  415. <a href="https://agentfield.ai/github/pr-af/?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-pr-af-repo">View project →</a>
  416. </td>
  417. </tr>
  418. <tr>
  419. <td align="center" width="50%">
  420. <a href="https://github.com/Agent-Field/person-to-brief-af">
  421. <img src="assets/examples/person-to-brief-af.png" alt="Person to Brief AF" />
  422. </a>
  423. <br/>
  424. <b>Person to Brief AF</b>
  425. <br/>
  426. <sub>Give it a name and company. It resolves the right identity, researches public professional sources in parallel, verifies claims, and returns an outreach-ready brief plus cited PDF.</sub>
  427. <br/><br/>
  428. <a href="https://github.com/Agent-Field/person-to-brief-af">View project →</a>
  429. </td>
  430. <td align="center" width="50%">
  431. </td>
  432. </tr>
  433. </table>
  434. [See all examples →](https://agentfield.ai/examples?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-see-all-examples)
  435. Built something with AgentField? [Submit your project to be featured on the examples page](https://github.com/Agent-Field/agentfield/issues/new?template=community-project.md).
  436. ## Architecture
  437. <div align="center">
  438. <img src="assets/arch.png" alt="AgentField Architecture" width="100%" />
  439. </div>
  440. The control plane is a stateless Go service. Agents connect from anywhere - your laptop, Docker, Kubernetes. They register capabilities, the control plane routes calls between them, tracks execution as DAGs, and enforces policies. [Full architecture docs →](https://agentfield.ai/docs/learn/architecture?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-architecture)
  441. ## Learn More
  442. AgentField sends privacy-minimized installation and lifecycle telemetry by default; no prompts or execution payloads are sent. Set `AGENTFIELD_TELEMETRY_ENABLED=false` to disable it. [Telemetry details](docs/ENVIRONMENT_VARIABLES.md#anonymous-telemetry).
  443. The thinking behind AgentField - essays on AI backends, harness orchestration, and the infrastructure production agents actually need.
  444. <table>
  445. <tr>
  446. <td align="center" width="50%">
  447. <a href="https://agentfield.ai/blog/what-is-harness-orchestration?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-blog-what-is-harness-orchestration">
  448. <img src="assets/blog/what-is-harness-orchestration.webp" alt="What is harness orchestration?" />
  449. </a>
  450. <br/>
  451. <b>What is harness orchestration?</b>
  452. <br/>
  453. <sub>The atomic unit of intelligence is climbing from the model call to the autonomous harness - and what changes when it does.</sub>
  454. <br/><br/>
  455. <a href="https://agentfield.ai/blog/what-is-harness-orchestration?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-blog-what-is-harness-orchestration">Read post →</a>
  456. </td>
  457. <td align="center" width="50%">
  458. <a href="https://agentfield.ai/blog/harness-as-black-box?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-blog-harness-black-box">
  459. <img src="assets/blog/harness-black-box.webp" alt="Part 1: The Black Box" />
  460. </a>
  461. <br/>
  462. <b>Part 1: The Black Box</b>
  463. <br/>
  464. <sub>Treating harnesses like Claude Code and Codex as autonomous, embodied, persistent computational entities.</sub>
  465. <br/><br/>
  466. <a href="https://agentfield.ai/blog/harness-as-black-box?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-blog-harness-black-box">Read post →</a>
  467. </td>
  468. </tr>
  469. <tr>
  470. <td align="center" width="50%">
  471. <a href="https://agentfield.ai/blog/harness-as-membrane?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-blog-harness-membrane">
  472. <img src="assets/blog/harness-membrane.webp" alt="Part 2: Engineering the Membrane" />
  473. </a>
  474. <br/>
  475. <b>Part 2: Engineering the Membrane</b>
  476. <br/>
  477. <sub>Shaping the boundary surface of a harness across four engineerable dimensions: workspace, drift, verifier placement, and recovery budget.</sub>
  478. <br/><br/>
  479. <a href="https://agentfield.ai/blog/harness-as-membrane?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-blog-harness-membrane">Read post →</a>
  480. </td>
  481. <td align="center" width="50%">
  482. <a href="https://agentfield.ai/blog/ai-backend?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-blog-ai-backend">
  483. <img src="assets/blog/ai-backend.webp" alt="The AI Backend" />
  484. </a>
  485. <br/>
  486. <b>The AI Backend</b>
  487. <br/>
  488. <sub>Our thesis: in five years every serious software company will run an AI backend - a reasoning layer that makes the decisions that used to be hardcoded.</sub>
  489. <br/><br/>
  490. <a href="https://agentfield.ai/blog/ai-backend?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-blog-ai-backend">Read post →</a>
  491. </td>
  492. </tr>
  493. <tr>
  494. <td align="center" width="50%">
  495. <b>Fan out 1,000 parallel agents from one request</b>
  496. <br/>
  497. <sub>A tutorial on turning a single call into a bounded fan-out across the control plane, with queues, retries, and traces on every branch.</sub>
  498. <br/><br/>
  499. <a href="https://agentfield.ai/blog/fan-out-1000-agents?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-blog-fan-out-1000-agents">Read post →</a>
  500. </td>
  501. <td align="center" width="50%">
  502. <b>Claude Code as a function</b>
  503. <br/>
  504. <sub>Wrap a multi-turn coding harness behind a REST endpoint and call Claude Code the same way you call any other agent.</sub>
  505. <br/><br/>
  506. <a href="https://agentfield.ai/blog/claude-code-as-a-function?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-blog-claude-code-as-a-function">Read post →</a>
  507. </td>
  508. </tr>
  509. </table>
  510. ### Documentation
  511. - **[vs Agent Frameworks](https://agentfield.ai/docs/learn/vs-frameworks?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-vs-frameworks)** - How AgentField compares to LangChain, CrewAI, and workflow engines
  512. - **[Full Documentation](https://agentfield.ai/docs/learn?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-full-docs)**
  513. ## New from AgentField: CodeAF
  514. <a href="https://agentfield.ai/github/codeaf/?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-codeaf-repo">
  515. <img src="https://raw.githubusercontent.com/Agent-Field/CodeAF/dev/assets/readme/benchmark-deepswe.webp" alt="First on DeepSWE: senior-dev, CodeAF's developer subharness, solved the most tasks (54.9%) at the lowest cost per solved task (1x). Every other harness solved less and paid more per solve: mini-swe-agent 1.9x, codex 2.1x, pi 2.4x, claude-code 3.4x, omp, kilo and opencode about 4.5x, muse-code 11.3x, deepseek-harness 26.6x." width="100%">
  516. </a>
  517. **[CodeAF](https://agentfield.ai/github/codeaf/?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-codeaf-repo)** is our new open-source coding harness,
  518. built for open models: frontier-grade coding at a fraction of the cost. It ranked
  519. **#1 on DeepSWE** out of ten coding harnesses on the same model (DeepSeek V4 Flash),
  520. ahead of Claude Code, Codex and OpenCode, at the lowest cost per solved issue.
  521. One Go binary, Apache 2.0.
  522. [Try CodeAF →](https://agentfield.ai/github/codeaf/?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-codeaf-repo) · [See the benchmark →](https://agentfield.ai/github/codeaf/?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-codeaf-benchmark#benchmarks)
  523. ## Community
  524. <div align="center">
  525. [![Discord](https://img.shields.io/badge/Join%20Discord-d4a24a?style=for-the-badge&logo=discord&logoColor=white)](https://discord.gg/aBHaXMkpqh)
  526. [![Twitter](https://img.shields.io/badge/Follow%20on%20X-0c0b09?style=for-the-badge&logo=x&logoColor=white)](https://x.com/agentfield_ai)
  527. **[GitHub Issues](https://github.com/Agent-Field/agentfield/issues)** · **[Documentation](https://agentfield.ai/docs/learn?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-community-docs)** · **[Examples](https://agentfield.ai/docs/learn/examples?utm_source=github-readme&utm_campaign=github-readme&utm_id=github-readme-community-examples)**
  528. </div>
  529. ## License
  530. [Apache 2.0](LICENSE)

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.