Coworker API
Build agents on your company's context graph.
The Coworker API connects your context layer to your own agents and applications. Every call draws on permission-aware context graphs from 50+ of your tools, so agents are more accurate, reliable and cheaper to run.
Trusted by
84.5%
better quality, preferred in blind review
64%
faster to complete a task
9x
cheaper per answer
Statistically significant benchmarks comparing Coworker MCP vs. 3rd-party MCPs in Claude
The problem
Your internal agents are only as reliable as the context they can reach.
The solution · OM2
OM2 is the context graph behind every call.
OM2 links the facts across your company's tools, such as customers, projects, people and decisions, and keeps them current, traced to their source and scoped to each person's access. Agents stop guessing because every fact has a source. They stop stalling because they can see who owns what and what changed. And they stop paying to re-read whole threads, because they read the graph instead.
Mosaic Financial
Series B fintech · $3.1M ARR · 47 seats
project
SOC 2 Type II audit
person
Priya Nambiar
workflow
Enterprise onboarding
team
Customer Success
source
#mosaic-impl
Mosaic Financial
Series B fintech · $3.1M ARR · 47 seats
Use cases
Built for the agents and apps you're already shipping.
Autonomous agents
Give agents built in LangGraph, CrewAI or any SDK the context a senior employee has. They act as their person, and nothing goes out without approval.
Your own enterprise AI app
Build a branded chat app or in-product assistant. Coworker handles retrieval, model choice and grounding, and threads sync to the Coworker app.
Enterprise search
Add search over your connected tools to your intranet or help desk. Results respect each person's permissions, and search is read-only by default.
Three ways to call Coworker
Bring your own agent, or use ours.
One API, three ways in. Pick the one that matches what you're building.
Your agent reasons
Call Coworker from your own agent.
Call any of the one-off agents in your Coworker workspace as a tool, or run a graph search over your company's context directly. Definitions come back in MCP, OpenAI or Anthropic format, so they drop into the SDK you already use.
- GET /v1/tools?format=openai
- POST /v1/tools/{name}/call
tools = requests.get(f"{BASE}/v1/tools",
params={"format": "openai"}, headers=H).json()["data"]
resp = client.chat.completions.create(
model="gpt-4.1", messages=messages, tools=tools)
for call in resp.choices[0].message.tool_calls:
result = requests.post(
f"{BASE}/v1/tools/{call.function.name}/call",
headers=H,
json={"arguments": json.loads(call.function.arguments)})Security and governance
Your context and your stack stay yours.
Built for enterprises that want to own their AI stack: you keep the graph, choose the models and control who sees what, hosted by Coworker or in your own cloud.
Every agent sees exactly what its person can see.
Scoped keys
Eight scopes, none implied by another. A key does exactly what it was minted with.
Act on behalf of a member
Service keys name the member on every request, and acting for others needs an admin-created key.
Read-only by default
Without write scope, agents can search and read but are never offered a tool that changes anything.
Approval before action
Writes from an agent run land as drafts until the member approves them.
Audit sessions
Group related calls with a session id so every agent action is traceable.
Private by member
Conversations, runs and drafts are visible only to the member a request acts as.
Compare
Three ways to give your agents a context graph.
| Feature | Coworker API | Data-platform context APIs | Build it yourself |
|---|---|---|---|
| Context graph | Built and kept current for you: facts about people, deals, meetings, tickets, docs and code, linked across 50+ tools | Metadata and semantic layers over warehouse tables, curated by data stewards | Yours to design, build and maintain: connectors, entity linking, freshness |
| Time to first answer | As soon as your network is set up and your tools are connected | Weeks of modeling and curation | Months of connectors and pipelines |
| Permissions | Per member, enforced on every call | Inherited from the data platform | Built and maintained per agent |
| Actions | Drafts with approval, by API or in the app | Limited or platform-specific | Custom code |
| Model choice | Any model in your loop, or routed for you | Usually the platform's | Yours |
Frequently asked questions
An API that lets your own agents and applications query a context graph. An agent asks a question or calls a tool and gets back linked company facts with their sources, limited to what the person it acts for is allowed to see. The Coworker API is Coworker's context graph API.
Give every agent the context your people have.
See the API on your own data in a 20-minute demo. From there, we'll get your network set up so your team can start building.