For developers
The AI feature is due this sprint. Retrieval, sandboxes, and auth are still on the whiteboard.
Start a chat with one of your Agents from your own code, over the REST API.
No credit card required.
$ curl -X POST https://app.revelry.ai/api/v2/chats \
-H "Authorization: Bearer $REVELRY_API_KEY" \
-H "Content-Type: application/json" \
-d '{"project_id": 1, "agent_id": 7, "content": "Which subs are late on submittals?"}'
The primitives
Three primitives you can build on today
Skills
A Skill is a prompt with an optional model configuration, tools, and data sources. Skills can call other Skills, either in a fixed order or when the model decides it needs one. You run a Skill over the API, or an Agent runs it as a tool.
Agents
An Agent is a configured chat: a persona, memory, Skills, tools, and data sources. Set one up in the app, or describe what you need in chat and Revelry Ai drafts it.
Workflows
A Workflow is a multi-step pipeline with explicit inputs and outputs between steps. Use one when a process has to run the same way every time, and start runs over the API.
Kits will package Skills, Agents, and Workflows for one industry. They aren't in the product yet.
Use cases
What you can build with this
Document processing pipelines
Chain extraction, classification, and validation Skills. Send contracts, invoices, or applications over the API, then collect structured data when the run finishes.
3 Skills, run over the API
Agents that work in GitHub
Connect GitHub with a token and an Agent gets GitHub's MCP tools in chat. It can read issues, look through code, and open pull requests for your team to review and merge.
GitHub MCP tools, with approval on every write
Support Agents grounded in your docs
Add your docs and FAQs to Knowledge. The Agent answers from them and cites the source for each answer, and your product can send it questions through the chat API.
Knowledge and an Agent, called through /api/v2/chats
Multi-step analysis Workflows
Run intake, analysis, and report generation as steps in one Workflow. Each step can use a different model, and each step's output feeds the next.
A Workflow of 3 steps, ending in an Artifact
MCP
MCP works in both directions
As a server
Outside MCP clients, such as Claude, Cursor, or your own apps, can call your Revelry Ai Agents and Skills directly. Build a capability once in Revelry Ai, and every client you use can reach it.
Built-in tools for your Agents
Every Agent can call tools for your Knowledge, the web, Artifacts, Skills, and Workspaces. The Skill tools let an Agent find, run, and write Skills:
skill_list
skill_get
skill_execute
skill_create
skill_update
Outside MCP servers
Connect an MCP server, such as GitHub's, and its tools are available in chat and in Workflow runs. Read-only tools run on their own. Anything that creates, updates, or deletes waits for someone to approve it.
Workspaces
Sandboxes where Agents run code
Each Workspace is an isolated sandbox with one of four profiles. Your Agent drives it through tool calls, and the reasoning stays in the Agent.
| Profile | What it's set up for |
|---|---|
| document | Python with pandas, openpyxl, LibreOffice Calc, and PDF tools, for working through documents |
| data_analysis | The document environment, used for analysis and charts |
| spreadsheet | Reading, building, and updating spreadsheets, with openpyxl, xlsxwriter, and pandas installed |
| general | A base sandbox for anything else |
What a Workspace gives an Agent
- Shell commands and code execution
- Files it can read, write, search, and edit
- Tools to analyze, create, and update spreadsheets
- Isolation from your own systems, so a bad command stays in the sandbox
Your Agent decides what runs
The sandbox executes commands, and your Agent chooses which ones, using these tools:
API
Start a chat with an Agent, then keep talking.
Create a chat with an Agent and your first message, and the Agent's answer comes back in the same response. Send more messages to the same chat to continue. For long answers, pass "stream": true and read the reply as it streams.
start the chat
POST /api/v2/chats
Authorization: Bearer $REVELRY_API_KEY
Content-Type: application/json
{
"project_id": 1,
"agent_id": 7,
"content": "Which subs are late on submittals?"
}
201 Created
{
"status": "ok",
"response": {
"id": 42,
"agent_id": 7,
"message": {
"role": "assistant",
"content": "Three subs are past due...",
"tool_executions": [...]
}
}
}
keep the conversation going
POST /api/v2/chats/42/messages
Authorization: Bearer $REVELRY_API_KEY
Content-Type: application/json
{
"content": "Draft a reminder to each of them."
}
200 OK
{
"status": "ok",
"response": {
"message": {
"role": "assistant",
"content": "Here are three drafts...",
"tool_executions": []
}
}
}
When you need one structured result instead of a conversation, run a Skill directly. Skill runs are asynchronous: the POST returns a job ID, and you poll the job for the result.
POST /api/v2/chats # Start a chat, optionally with an Agent
POST /api/v2/chats/:id/messages # Send a message to a chat
GET /api/v2/chats/:id/messages # Read a chat's history
POST /api/v2/skills/:skill_id/run # Start a Skill run
GET /api/v2/api_jobs/:id # Poll a Skill run for its result
GET /api/v1/skills # List Skills
POST /api/v1/workflows/:workflow_id/runs # Start a Workflow run
Retrieval
Retrieval is built in
Add documents and data sources to Knowledge, group them into collections, and point a Skill or Agent at them. Revelry Ai handles the chunking, embedding, storage, and search.
Models
Use the right model for the job
Set the model per Skill or per Agent, and pass a model configuration with an API run to override it. Revelry Ai works with models from Anthropic, OpenAI, Google, Mistral, and xAI. Enterprise customers can bring their own API keys.
Bring the first Agent you want to call.
Sign up, create an API key, and start a chat with it from your own code the same day.
No credit card required.
Talk to the people who built it.
Email dev@revelry.ai with what you're building. An engineer on the platform team answers.
Questions developers ask
You can, and Revelry Ai calls them too. A raw model call gives you a prompt and a response. A product also needs retrieval over your data, tool calls, sandboxes, multi-step runs, auth, audit trails, and a way to change models without rewriting your app. Revelry Ai runs that layer, so your code starts a chat with an Agent and gets the answer.
In the app, with the skill_create MCP tool, or by describing what you need in chat. Revelry Ai drafts it for you to review.
Yes. A Skill can call other Skills, and an Agent can hand part of a task to a sub-agent.
Not yet. Today you call the REST API, or connect any MCP client to your Agents and Skills.
Managed VPC deployments are available on enterprise contracts. We are also looking at a license for self-hosted deployments. If you need one, tell us.
Elixir, on the Phoenix framework. We chose it for its concurrency model and its fault tolerance, which suit AI work that runs for minutes at a time instead of milliseconds.
Email dev@revelry.ai. An engineer on the platform team answers.