AI-gent Workflows

# AI-gent Workflows (secai)

is a stateful layer between user prompts and LLMs

**Demo 2** - recipe cooking agent

- agent has several prompts
- and a **state machine**
- consensus between the prompts
- each agent has stories
- with actors
- agent can create **state machines**
- using LLMs
- _dynamic short-term memory_
- non-actionable
- eg `IngredientCarrots`
- actionable
- eg `StepBakeCake`
- long-term memory is in **SQLite**
- 1 file with all the prompts

**Demo 2** - recipe cooking agent

- every action can be explained
- total control over the flow
- reasoning happens mostly
- inside the _state machine_
- devtools feature
- time travel
- visualizations
- graph navigation
- telemetry
- agents can communicate
- _statefully_ over the network
- live REPL can
- alter memory
- alter agents

**Demo 2** - recipe cooking agent

- agent remembers jokes
- how many vs
- the number of completed steps
- user can ask the agent to
- switch to another story
- complete a cooking step
- some prompts happen in parallel
- cooking steps are a DAG
- all content is AI generated
- agent has an easy schema
- states
- prompts
- stories
- see at /files/schema.go

**Demo 2** - recipe cooking agent

- each agent has many UIs
- each UI can have many instances
- each UI shows the same agent
- pure vector graphics
- Web Browser transport provided
- canvas-based
- target platforms
- SSH
- Web
- mobile Web
- desktops
- native iOS
- native Android
- on-device deployments
- privacy focused

### Visit https:// ai-gents.work
### FOSS https:// github.com/pancsta/secai

**AI-gent Workflows** (aka **secai**) is a workflow orchestrator for
**autonomous LLM bots**. It's implemented on top of a **unified
state graph** and makes a solid foundation for complex, proactive and
long-lived **AI Agents** with deep and structured memory. By having graph-based flow, **secai** allows for precise behavior modeling of agents, including interruptions and fault-tolerancy.

Roadmap

- MCP
- history DSL with a vector format (WIP)
- contracts between agents
- i18n
- ML-based triggers (local)
- Gemini

### Thank You