agentcogs 0.1.1__tar.gz → 0.1.3__tar.gz

agentcogs-0.1.3/PKG-INFO

@@ -0,0 +1,115 @@
+Metadata-Version: 2.5
+Name: agentcogs
+Version: 0.1.3
+Summary: Customer-level AI unit economics. Compare each customer's LLM cost with their monthly fee
+Project-URL: Homepage, https://agentcogs.vercel.app
+Project-URL: Repository, https://github.com/vaibhav11123/agentcogs
+Project-URL: Issues, https://github.com/vaibhav11123/agentcogs/issues
+Author-email: AgentCOGS <hello@agentcogs.dev>
+License: MIT
+License-File: LICENSE
+Keywords: anthropic,attribution,billing,cost,langchain,langgraph,llm,margin,openai,tenant,unit-economics
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Topic :: Software Development :: Libraries
+Requires-Python: >=3.10
+Requires-Dist: httpx>=0.27.0
+Requires-Dist: shekel[anthropic,openai]<2,>=1.2
+Provides-Extra: dev
+Requires-Dist: pytest-asyncio>=0.23; extra == 'dev'
+Requires-Dist: pytest>=8.0; extra == 'dev'
+Requires-Dist: respx>=0.20; extra == 'dev'
+Requires-Dist: ruff>=0.6; extra == 'dev'
+Requires-Dist: starlette>=0.27.0; extra == 'dev'
+Provides-Extra: integrations
+Requires-Dist: starlette>=0.27.0; extra == 'integrations'
+Description-Content-Type: text/markdown
+
+# AgentCOGS
+
+AgentCOGS is an open-source Python SDK and a self-hosted dashboard for customer-level AI unit economics. It attributes each LLM call to the customer who caused it, compares that cost with the monthly fee, and checks a per-customer budget before the next SDK call.
+
+**Know which AI customers make you money, and which ones are eating your margins.** The blended total can look fine while one fee no longer covers the AI cost. Which customer stops making money this month.
+
+```bash
+pip install agentcogs
+```
+
+This package is the SDK. It is on PyPI. The customer list is the dashboard in the [repository](https://github.com/vaibhav11123/agentcogs), and you run that yourself. A hosted API is not serving. The working sample is [agentcogs.vercel.app](https://agentcogs.vercel.app).
+
+## What does the number mean?
+
+AI margin is `(monthly fee − attributed AI cost) / monthly fee`. It is the contribution of the fee after attributed AI cost. It is not total customer profitability, and it is not accounting gross margin. People, infrastructure, support, payment processing, refunds, discounts, credits, and proration are outside it. Pace is a straight line from the cost so far. It does not forecast a change in behavior. You type the fee, or you pull that customer's Stripe subscription. The SDK does not invent it.
+
+## Install, configure, instrument, verify
+
+`AGENTCOGS_ENDPOINT` defaults to `http://localhost:8000`. Create a workspace with the self-hosted dashboard, then:
+
+```python
+import os
+from openai import OpenAI
+import agentcogs
+
+agentcogs.init()  # AGENTCOGS_API_KEY, AGENTCOGS_WORKSPACE_ID, AGENTCOGS_ENDPOINT
+print(agentcogs.ping())
+agentcogs.set_customer("acme")  # your tenant id, once per request
+
+with agentcogs.run(workflow_id="support") as ctx:
+    if os.environ.get("OPENAI_API_KEY"):
+        OpenAI().chat.completions.create(
+            model="gpt-4o-mini",
+            messages=[{"role": "user", "content": "Say hi in one word."}],
+            max_tokens=5,
+        )
+
+print(ctx.wait_for_ingest().accepted, ctx.run_id)
+```
+
+The SDK does not call a model. Inside the block, shekel reads token usage from the OpenAI or Anthropic call you make. When the block exits, that cost is posted. With no `OPENAI_API_KEY`, the same block posts a zero-cost run so you can see the row. Open that customer and type the monthly fee.
+
+`python examples/hello_agentcogs.py` in the repository is this path. `python examples/one_run.py` posts one recorded cost of USD 0.84 and does not call a model.
+
+## How attribution works
+
+`set_customer()` names the tenant for this request. `run()` watches the instrumented client calls in the block and posts one event with `total_usd` when the block exits. If no customer is set, `run()` raises before the block. A duplicate `run_id` is accepted as a duplicate and is not stored twice. A failed post sits in a local outbox and is retried. `python -m agentcogs outbox status` shows what has not been delivered.
+
+## What the budget check does
+
+If `monthly_budget_usd` is already exceeded, the next `run()` raises `CustomerBudgetExceededError` and does not enter the block. Two calls already in flight can both pass. It is not an atomic reservation. A LiteLLM callback records cost after the provider has already answered, so it cannot stop that call.
+
+LiteLLM does not ship the callback. You copy it from the repository. The FastAPI middleware is a separate extra: `pip install "agentcogs[integrations]"`.
+
+## Supported integrations
+
+| | |
+| --- | --- |
+| OpenAI and Anthropic | Usage read inside `run()` by shekel, which this package depends on |
+| [FastAPI](https://github.com/vaibhav11123/agentcogs/blob/main/docs/integrations/fastapi.md) | Set the tenant from the request. Needs the `integrations` extra |
+| [LangGraph](https://github.com/vaibhav11123/agentcogs/blob/main/docs/integrations/langgraph.md) | Wrap the graph |
+| [LiteLLM](https://github.com/vaibhav11123/agentcogs/blob/main/docs/integrations/litellm.md) | A callback you register. It posts after the proxy call |
+| Stripe | The subscription is the fee. An optional meter pushes usage you already attributed |
+| [Slack](https://github.com/vaibhav11123/agentcogs/blob/main/docs/integrations/slack.md) | A spike opens that customer |
+
+## One fictional account
+
+[Patternstack](https://agentcogs.vercel.app) is a snapshot of forty fictional customers, as of 14 Nov 2026. The blended AI margin is 74.6%. Vellum Research has spent USD 5,822.84 against a fee of USD 8,200, so it is 29.0% so far, and the straight-line pace ends near −52.2%. The sample does not receive your calls.
+
+<img src="https://raw.githubusercontent.com/vaibhav11123/agentcogs/main/docs/assets/patternstack.gif" alt="Opening the fictional Patternstack sample. AI margin is 74.6%. Vellum Research is 29.0% so far and the month ends near −52.2%." width="840">
+
+## Documentation
+
+- [Quickstart](https://github.com/vaibhav11123/agentcogs/blob/main/docs/quickstart.md)
+- [Self-host](https://github.com/vaibhav11123/agentcogs/blob/main/docs/DEPLOY.md)
+- [Customer id](https://github.com/vaibhav11123/agentcogs/blob/main/docs/concepts/customer-id.md)
+- [Docs index](https://github.com/vaibhav11123/agentcogs/blob/main/docs/README.md)
+- [Repository README](https://github.com/vaibhav11123/agentcogs/blob/main/README.md)
+- Sample: https://agentcogs.vercel.app
+- Source: https://github.com/vaibhav11123/agentcogs
+
+## What is not guaranteed
+
+A match against the provider invoice, refunds, credits, discounts, currency conversion, and a hard stop when two calls pass the pre-check together. Development status is Beta: the SDK surface is `init`, `set_customer`, `run`, `ping`, and `shutdown`, and a hosted API is not part of this release.

agentcogs-0.1.3/PYPI.md

@@ -0,0 +1,84 @@
+# AgentCOGS
+
+AgentCOGS is an open-source Python SDK and a self-hosted dashboard for customer-level AI unit economics. It attributes each LLM call to the customer who caused it, compares that cost with the monthly fee, and checks a per-customer budget before the next SDK call.
+
+**Know which AI customers make you money, and which ones are eating your margins.** The blended total can look fine while one fee no longer covers the AI cost. Which customer stops making money this month.
+
+```bash
+pip install agentcogs
+```
+
+This package is the SDK. It is on PyPI. The customer list is the dashboard in the [repository](https://github.com/vaibhav11123/agentcogs), and you run that yourself. A hosted API is not serving. The working sample is [agentcogs.vercel.app](https://agentcogs.vercel.app).
+
+## What does the number mean?
+
+AI margin is `(monthly fee − attributed AI cost) / monthly fee`. It is the contribution of the fee after attributed AI cost. It is not total customer profitability, and it is not accounting gross margin. People, infrastructure, support, payment processing, refunds, discounts, credits, and proration are outside it. Pace is a straight line from the cost so far. It does not forecast a change in behavior. You type the fee, or you pull that customer's Stripe subscription. The SDK does not invent it.
+
+## Install, configure, instrument, verify
+
+`AGENTCOGS_ENDPOINT` defaults to `http://localhost:8000`. Create a workspace with the self-hosted dashboard, then:
+
+```python
+import os
+from openai import OpenAI
+import agentcogs
+
+agentcogs.init()  # AGENTCOGS_API_KEY, AGENTCOGS_WORKSPACE_ID, AGENTCOGS_ENDPOINT
+print(agentcogs.ping())
+agentcogs.set_customer("acme")  # your tenant id, once per request
+
+with agentcogs.run(workflow_id="support") as ctx:
+    if os.environ.get("OPENAI_API_KEY"):
+        OpenAI().chat.completions.create(
+            model="gpt-4o-mini",
+            messages=[{"role": "user", "content": "Say hi in one word."}],
+            max_tokens=5,
+        )
+
+print(ctx.wait_for_ingest().accepted, ctx.run_id)
+```
+
+The SDK does not call a model. Inside the block, shekel reads token usage from the OpenAI or Anthropic call you make. When the block exits, that cost is posted. With no `OPENAI_API_KEY`, the same block posts a zero-cost run so you can see the row. Open that customer and type the monthly fee.
+
+`python examples/hello_agentcogs.py` in the repository is this path. `python examples/one_run.py` posts one recorded cost of USD 0.84 and does not call a model.
+
+## How attribution works
+
+`set_customer()` names the tenant for this request. `run()` watches the instrumented client calls in the block and posts one event with `total_usd` when the block exits. If no customer is set, `run()` raises before the block. A duplicate `run_id` is accepted as a duplicate and is not stored twice. A failed post sits in a local outbox and is retried. `python -m agentcogs outbox status` shows what has not been delivered.
+
+## What the budget check does
+
+If `monthly_budget_usd` is already exceeded, the next `run()` raises `CustomerBudgetExceededError` and does not enter the block. Two calls already in flight can both pass. It is not an atomic reservation. A LiteLLM callback records cost after the provider has already answered, so it cannot stop that call.
+
+LiteLLM does not ship the callback. You copy it from the repository. The FastAPI middleware is a separate extra: `pip install "agentcogs[integrations]"`.
+
+## Supported integrations
+
+| | |
+| --- | --- |
+| OpenAI and Anthropic | Usage read inside `run()` by shekel, which this package depends on |
+| [FastAPI](https://github.com/vaibhav11123/agentcogs/blob/main/docs/integrations/fastapi.md) | Set the tenant from the request. Needs the `integrations` extra |
+| [LangGraph](https://github.com/vaibhav11123/agentcogs/blob/main/docs/integrations/langgraph.md) | Wrap the graph |
+| [LiteLLM](https://github.com/vaibhav11123/agentcogs/blob/main/docs/integrations/litellm.md) | A callback you register. It posts after the proxy call |
+| Stripe | The subscription is the fee. An optional meter pushes usage you already attributed |
+| [Slack](https://github.com/vaibhav11123/agentcogs/blob/main/docs/integrations/slack.md) | A spike opens that customer |
+
+## One fictional account
+
+[Patternstack](https://agentcogs.vercel.app) is a snapshot of forty fictional customers, as of 14 Nov 2026. The blended AI margin is 74.6%. Vellum Research has spent USD 5,822.84 against a fee of USD 8,200, so it is 29.0% so far, and the straight-line pace ends near −52.2%. The sample does not receive your calls.
+
+<img src="https://raw.githubusercontent.com/vaibhav11123/agentcogs/main/docs/assets/patternstack.gif" alt="Opening the fictional Patternstack sample. AI margin is 74.6%. Vellum Research is 29.0% so far and the month ends near −52.2%." width="840">
+
+## Documentation
+
+- [Quickstart](https://github.com/vaibhav11123/agentcogs/blob/main/docs/quickstart.md)
+- [Self-host](https://github.com/vaibhav11123/agentcogs/blob/main/docs/DEPLOY.md)
+- [Customer id](https://github.com/vaibhav11123/agentcogs/blob/main/docs/concepts/customer-id.md)
+- [Docs index](https://github.com/vaibhav11123/agentcogs/blob/main/docs/README.md)
+- [Repository README](https://github.com/vaibhav11123/agentcogs/blob/main/README.md)
+- Sample: https://agentcogs.vercel.app
+- Source: https://github.com/vaibhav11123/agentcogs
+
+## What is not guaranteed
+
+A match against the provider invoice, refunds, credits, discounts, currency conversion, and a hard stop when two calls pass the pre-check together. Development status is Beta: the SDK surface is `init`, `set_customer`, `run`, `ping`, and `shutdown`, and a hosted API is not part of this release.

{agentcogs-0.1.1 → agentcogs-0.1.3}/pyproject.toml

@@ -5,12 +5,12 @@ build-backend = "hatchling.build"
 [project]
 name = "agentcogs"
 dynamic = ["version"]
-description = "Per-customer LLM cost attribution and runtime budget caps for agent SaaS"
-readme = "README.md"
+description = "Customer-level AI unit economics. Compare each customer's LLM cost with their monthly fee"
+readme = "PYPI.md"
 license = { text = "MIT" }
 requires-python = ">=3.10"
 authors = [{ name = "AgentCOGS", email = "hello@agentcogs.dev" }]
-keywords = ["llm", "cost", "attribution", "langgraph", "langchain", "openai", "anthropic", "billing"]
+keywords = ["llm", "cost", "attribution", "margin", "unit-economics", "tenant", "langgraph", "langchain", "openai", "anthropic", "billing"]
 classifiers = [
     "Development Status :: 4 - Beta",
     "Intended Audience :: Developers",
@@ -23,7 +23,6 @@ classifiers = [
 dependencies = [
     "shekel[openai,anthropic]>=1.2,<2",
     "httpx>=0.27.0",
-    "starlette>=0.27.0",
 ]
 
 [project.optional-dependencies]
@@ -31,7 +30,7 @@ dev = ["pytest>=8.0", "pytest-asyncio>=0.23", "respx>=0.20", "ruff>=0.6", "starl
 integrations = ["starlette>=0.27.0"]
 
 [project.urls]
-Homepage = "https://agentcogs.dev"
+Homepage = "https://agentcogs.vercel.app"
 Repository = "https://github.com/vaibhav11123/agentcogs"
 Issues = "https://github.com/vaibhav11123/agentcogs/issues"
 
@@ -44,7 +43,7 @@ packages = ["src/agentcogs"]
 # pip install prefers the wheel. The sdist is the fallback and must stay the
 # SDK. The backend, dashboard, and demo tools are the rest of this repo.
 [tool.hatch.build.targets.sdist]
-only-include = ["src/agentcogs", "README.md", "LICENSE"]
+only-include = ["src/agentcogs", "PYPI.md", "LICENSE"]
 
 [tool.ruff]
 line-length = 100

{agentcogs-0.1.1 → agentcogs-0.1.3}/src/agentcogs/__init__.py

@@ -2,7 +2,7 @@
 
 Quickstart:
     import agentcogs
-    agentcogs.init(api_key="acg_live_...", workspace_id="...")
+    agentcogs.init()  # AGENTCOGS_API_KEY, AGENTCOGS_WORKSPACE_ID, AGENTCOGS_ENDPOINT
 
     agentcogs.set_customer("tenant_42")  # once per request
     with agentcogs.run(workflow_id="summarize"):

agentcogs-0.1.3/src/agentcogs/_version.py

@@ -0,0 +1 @@
+__version__ = "0.1.3"

{agentcogs-0.1.1 → agentcogs-0.1.3}/src/agentcogs/config.py

@@ -12,7 +12,7 @@ BudgetMode = Literal["open", "closed"]
 class _Config:
     api_key: Optional[str] = None
     workspace_id: Optional[str] = None
-    endpoint: str = "https://api.agentcogs.dev"
+    endpoint: str = "http://localhost:8000"
     timeout_seconds: float = 2.0
     offline: bool = False
     budget_mode: BudgetMode = "open"
@@ -51,7 +51,7 @@ def init(
     _config.endpoint = (
         endpoint
         or os.environ.get("AGENTCOGS_ENDPOINT")
-        or "https://api.agentcogs.dev"
+        or "http://localhost:8000"
     ).rstrip("/")
     _config.offline = offline or os.environ.get("AGENTCOGS_OFFLINE") == "1"
     _config.timeout_seconds = timeout_seconds

agentcogs-0.1.1/PKG-INFO

@@ -1,149 +0,0 @@
-Metadata-Version: 2.5
-Name: agentcogs
-Version: 0.1.1
-Summary: Per-customer LLM cost attribution and runtime budget caps for agent SaaS
-Project-URL: Homepage, https://agentcogs.dev
-Project-URL: Repository, https://github.com/vaibhav11123/agentcogs
-Project-URL: Issues, https://github.com/vaibhav11123/agentcogs/issues
-Author-email: AgentCOGS <hello@agentcogs.dev>
-License: MIT
-License-File: LICENSE
-Keywords: anthropic,attribution,billing,cost,langchain,langgraph,llm,openai
-Classifier: Development Status :: 4 - Beta
-Classifier: Intended Audience :: Developers
-Classifier: License :: OSI Approved :: MIT License
-Classifier: Programming Language :: Python :: 3.10
-Classifier: Programming Language :: Python :: 3.11
-Classifier: Programming Language :: Python :: 3.12
-Classifier: Topic :: Software Development :: Libraries
-Requires-Python: >=3.10
-Requires-Dist: httpx>=0.27.0
-Requires-Dist: shekel[anthropic,openai]<2,>=1.2
-Requires-Dist: starlette>=0.27.0
-Provides-Extra: dev
-Requires-Dist: pytest-asyncio>=0.23; extra == 'dev'
-Requires-Dist: pytest>=8.0; extra == 'dev'
-Requires-Dist: respx>=0.20; extra == 'dev'
-Requires-Dist: ruff>=0.6; extra == 'dev'
-Requires-Dist: starlette>=0.27.0; extra == 'dev'
-Provides-Extra: integrations
-Requires-Dist: starlette>=0.27.0; extra == 'integrations'
-Description-Content-Type: text/markdown
-
-# AgentCOGS
-
-**Which customer stops making money this month.**
-
-You sell an agent to other companies. The model provider sends one invoice. AgentCOGS splits that spend by the tenant who caused it, sets it against the monthly fee, and shows where the month is headed if the calls keep coming.
-
-[![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](LICENSE)
-[![Python 3.10+](https://img.shields.io/badge/python-3.10+-blue.svg)](https://www.python.org/downloads/)
-[![CI](https://github.com/vaibhav11123/agentcogs/actions/workflows/ci.yml/badge.svg)](https://github.com/vaibhav11123/agentcogs/actions/workflows/ci.yml)
-
-The sample company is [Patternstack](https://agentcogs.vercel.app). One click, nothing to type. It is a snapshot of a fictional book of business, shipped in the dashboard, so it opens when no API is running. A hosted AgentCOGS API is not serving. To run your own customers, self-host.
-
----
-
-## The book, and the account inside it
-
-Patternstack, as of 14 Nov 2026. Forty customers. The month so far looks healthy.
-
-| | So far | If the calls continue |
-| --- | ---: | ---: |
-| AI cost | $39,613.25 | $84,885.54 |
-| Fees | $155,918.00 | $155,918.00 |
-| Margin | 74.6% | 45.6% |
-
-The fee is the monthly amount on the account. Pace is a straight line from the cost so far. It does not forecast a change in behavior.
-
-Two accounts are where the month breaks. Vellum Research has cost $5,822.84 against an $8,200 fee: **29.0%** so far, 2,724 runs, and the largest node is `synthesize`. If that spend continues, the month ends near **−52.2%**. Pinnacle Claims is 35.0% so far and ends near −39.3%. The blended 74.6% does not show either of those.
-
-<p align="center">
-  <img src="docs/assets/screenshots/patternstack-customers.png" alt="Patternstack customers. Blended margin 74.6%. Vellum Research is 29.0% and the month ends near −52.2%." width="880" />
-</p>
-
-Open the account and the same numbers sit next to the call that posted them.
-
-<p align="center">
-  <img src="docs/assets/screenshots/vellum-research.png" alt="Vellum Research. Margin 29.0%, AI cost $5,822.84, fee $8,200, and the set_customer call that posts each run." width="880" />
-</p>
-
-```python
-import agentcogs
-
-agentcogs.init()  # AGENTCOGS_API_KEY, AGENTCOGS_WORKSPACE_ID, AGENTCOGS_ENDPOINT
-agentcogs.set_customer("vellum_research")  # your tenant id, once per request
-
-with agentcogs.run(workflow_id="research_agent"):
-    result = agent.invoke(state)
-```
-
-The SDK does not call a model and does not invent the fee. It names the customer and posts the cost of the call you already made. You type the fee, or you pull the subscription from that customer's Stripe account. If no customer is set, the run stays unattributed.
-
-`python examples/one_run.py` posts one recorded $0.84 cost for Vellum Research so you can see a row land. It does not call a model.
-
----
-
-## Look, then run it on yours
-
-**Look.** [agentcogs.vercel.app](https://agentcogs.vercel.app) → **Open Patternstack**.
-
-**Run.** From this repo:
-
-```bash
-cp .env.selfhost.example .env
-docker compose up -d --build
-```
-
-Open http://localhost:3000 and choose **Create my workspace**. No email. The next screen has the install command with your key in it. The package is not on PyPI yet:
-
-```bash
-pip install "agentcogs @ git+https://github.com/vaibhav11123/agentcogs.git"
-```
-
-Point `AGENTCOGS_ENDPOINT` at `http://localhost:8000`. Send one call, open that customer, and type the monthly fee. Margin is the fee minus the cost.
-
-Full steps: [docs/quickstart.md](docs/quickstart.md). Self-host notes: [docs/DEPLOY.md](docs/DEPLOY.md).
-
----
-
-## Who it is for
-
-A founder of a product that sells one agent to many companies, priced as a monthly fee, with the model bill arriving as a single invoice. Support, research, claims, intake, document pipelines.
-
-| You already have | AgentCOGS adds |
-| --- | --- |
-| Langfuse or LangSmith | Which tenant the trace belonged to, and whether that tenant is still profitable |
-| LiteLLM | A callback that posts the tenant and the cost. The proxy stays |
-| Stripe | The subscription as the fee, and an optional meter for the usage you already attributed |
-| The provider invoice | The join from that one number to the customer list |
-
----
-
-## What you can connect
-
-| | |
-| --- | --- |
-| [FastAPI](docs/integrations/fastapi.md) | Set the tenant from the request |
-| [LangGraph](docs/integrations/langgraph.md) | Wrap the graph |
-| [LiteLLM](docs/integrations/litellm.md) | Post proxy traffic without wrapping the app |
-| [Customer import](docs/integrations/customer-import.md) | Load names and fees before the first run |
-| [Slack](docs/integrations/slack.md) | A spike opens that customer |
-| Stripe | Pull subscription fees. Push a daily usage meter |
-| Budget | `monthly_budget_usd` raises `CustomerBudgetExceededError` before the next call |
-
-Terminal proof for the SDK, the outbox, and a live call: [docs/demo-paths.md](docs/demo-paths.md).
-
----
-
-## Repo
-
-| Path | |
-| --- | --- |
-| `src/agentcogs/` | The SDK |
-| `backend/` | Ingest, margin, pace, budgets |
-| `dashboard/` | The book |
-| `examples/one_run.py` | One recorded cost, no model call |
-| `docs/` | [Index](docs/README.md) |
-
-Python SDK, FastAPI, Postgres, Redis, React. [MIT](LICENSE). [Contributing](CONTRIBUTING.md). [Security](SECURITY.md).

agentcogs-0.1.1/README.md

@@ -1,117 +0,0 @@
-# AgentCOGS
-
-**Which customer stops making money this month.**
-
-You sell an agent to other companies. The model provider sends one invoice. AgentCOGS splits that spend by the tenant who caused it, sets it against the monthly fee, and shows where the month is headed if the calls keep coming.
-
-[![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](LICENSE)
-[![Python 3.10+](https://img.shields.io/badge/python-3.10+-blue.svg)](https://www.python.org/downloads/)
-[![CI](https://github.com/vaibhav11123/agentcogs/actions/workflows/ci.yml/badge.svg)](https://github.com/vaibhav11123/agentcogs/actions/workflows/ci.yml)
-
-The sample company is [Patternstack](https://agentcogs.vercel.app). One click, nothing to type. It is a snapshot of a fictional book of business, shipped in the dashboard, so it opens when no API is running. A hosted AgentCOGS API is not serving. To run your own customers, self-host.
-
----
-
-## The book, and the account inside it
-
-Patternstack, as of 14 Nov 2026. Forty customers. The month so far looks healthy.
-
-| | So far | If the calls continue |
-| --- | ---: | ---: |
-| AI cost | $39,613.25 | $84,885.54 |
-| Fees | $155,918.00 | $155,918.00 |
-| Margin | 74.6% | 45.6% |
-
-The fee is the monthly amount on the account. Pace is a straight line from the cost so far. It does not forecast a change in behavior.
-
-Two accounts are where the month breaks. Vellum Research has cost $5,822.84 against an $8,200 fee: **29.0%** so far, 2,724 runs, and the largest node is `synthesize`. If that spend continues, the month ends near **−52.2%**. Pinnacle Claims is 35.0% so far and ends near −39.3%. The blended 74.6% does not show either of those.
-
-<p align="center">
-  <img src="docs/assets/screenshots/patternstack-customers.png" alt="Patternstack customers. Blended margin 74.6%. Vellum Research is 29.0% and the month ends near −52.2%." width="880" />
-</p>
-
-Open the account and the same numbers sit next to the call that posted them.
-
-<p align="center">
-  <img src="docs/assets/screenshots/vellum-research.png" alt="Vellum Research. Margin 29.0%, AI cost $5,822.84, fee $8,200, and the set_customer call that posts each run." width="880" />
-</p>
-
-```python
-import agentcogs
-
-agentcogs.init()  # AGENTCOGS_API_KEY, AGENTCOGS_WORKSPACE_ID, AGENTCOGS_ENDPOINT
-agentcogs.set_customer("vellum_research")  # your tenant id, once per request
-
-with agentcogs.run(workflow_id="research_agent"):
-    result = agent.invoke(state)
-```
-
-The SDK does not call a model and does not invent the fee. It names the customer and posts the cost of the call you already made. You type the fee, or you pull the subscription from that customer's Stripe account. If no customer is set, the run stays unattributed.
-
-`python examples/one_run.py` posts one recorded $0.84 cost for Vellum Research so you can see a row land. It does not call a model.
-
----
-
-## Look, then run it on yours
-
-**Look.** [agentcogs.vercel.app](https://agentcogs.vercel.app) → **Open Patternstack**.
-
-**Run.** From this repo:
-
-```bash
-cp .env.selfhost.example .env
-docker compose up -d --build
-```
-
-Open http://localhost:3000 and choose **Create my workspace**. No email. The next screen has the install command with your key in it. The package is not on PyPI yet:
-
-```bash
-pip install "agentcogs @ git+https://github.com/vaibhav11123/agentcogs.git"
-```
-
-Point `AGENTCOGS_ENDPOINT` at `http://localhost:8000`. Send one call, open that customer, and type the monthly fee. Margin is the fee minus the cost.
-
-Full steps: [docs/quickstart.md](docs/quickstart.md). Self-host notes: [docs/DEPLOY.md](docs/DEPLOY.md).
-
----
-
-## Who it is for
-
-A founder of a product that sells one agent to many companies, priced as a monthly fee, with the model bill arriving as a single invoice. Support, research, claims, intake, document pipelines.
-
-| You already have | AgentCOGS adds |
-| --- | --- |
-| Langfuse or LangSmith | Which tenant the trace belonged to, and whether that tenant is still profitable |
-| LiteLLM | A callback that posts the tenant and the cost. The proxy stays |
-| Stripe | The subscription as the fee, and an optional meter for the usage you already attributed |
-| The provider invoice | The join from that one number to the customer list |
-
----
-
-## What you can connect
-
-| | |
-| --- | --- |
-| [FastAPI](docs/integrations/fastapi.md) | Set the tenant from the request |
-| [LangGraph](docs/integrations/langgraph.md) | Wrap the graph |
-| [LiteLLM](docs/integrations/litellm.md) | Post proxy traffic without wrapping the app |
-| [Customer import](docs/integrations/customer-import.md) | Load names and fees before the first run |
-| [Slack](docs/integrations/slack.md) | A spike opens that customer |
-| Stripe | Pull subscription fees. Push a daily usage meter |
-| Budget | `monthly_budget_usd` raises `CustomerBudgetExceededError` before the next call |
-
-Terminal proof for the SDK, the outbox, and a live call: [docs/demo-paths.md](docs/demo-paths.md).
-
----
-
-## Repo
-
-| Path | |
-| --- | --- |
-| `src/agentcogs/` | The SDK |
-| `backend/` | Ingest, margin, pace, budgets |
-| `dashboard/` | The book |
-| `examples/one_run.py` | One recorded cost, no model call |
-| `docs/` | [Index](docs/README.md) |
-
-Python SDK, FastAPI, Postgres, Redis, React. [MIT](LICENSE). [Contributing](CONTRIBUTING.md). [Security](SECURITY.md).

agentcogs-0.1.1/src/agentcogs/_version.py

@@ -1 +0,0 @@
-__version__ = "0.1.1"