Rust-native agent system

UIntellAgent

One terminal. Full agency.

Chat, inspect, edit, execute, review, and remember without leaving the terminal. Every run keeps its context, every tool crosses a permission boundary, and every durable fact can live in a navigable graph.

Explore the workflow Run locally

Rust nativeone agent binary

SurrealDBgraph memory

Human gatedreviewed changes

UIntellAgent

provider ready

RUN / 01HF2AImplement graph-backed code context

verified

Inspect

Plan

Implement

Verify

Review

objective

Connect selected editor context to graph memory.

file_readsrc/editor.rs14 ms

file_writesrc/tools/graph.rsreview

code_execcargo test --all-targets95 passed

Change verified and ready for human review.

Unified TUIstate restored across sessions

Chatstream and intervene

Memoryinspect the graph

Toolsexecute by policy

Editorreview real files

Runsresume durable work

01 / WORKFLOW

Durable coding runs

A run is a process, not a disappearing reply.

UIntellAgent carries an objective through inspection, implementation, verification, independent review, and bounded repair. Checkpoints survive interruption and resume without replaying completed tool calls.

Give it an objective

The run captures the task, selected code context, workspace, provider, and explicit memory policy before work begins.

checkpoint created atomically

Inspect before editing

Files, symbols, prior runs, and graph context are gathered into a bounded working set. The plan remains visible beside tool events.

context ledger attached

Act through real tools

Shell state persists, file writes retain before-and-after bytes, and code execution uses Bubblewrap by default.

permission engine active

Verify, review, repair

Quality gates run against the workspace. Agent changes become independent hunks that can be accepted, rejected, or safely undone.

result ledger verified

02 / MEMORY

SurrealDB graph memory

Knowledge you can see, move, query, and correct.

Facts are not hidden in a transcript. They are inspectable units connected to datasets, code locations, decisions, and runs. Select and reposition the graph below.

X 0Y 0100%

Operate visually

Drag, pin, zoom, pan, lasso, filter, and auto-layout large knowledge graphs.

Edit relationships

Create and inspect typed edges, bulk-move units, and keep both directions visible.

Repair with history

Destructive graph actions are journaled, undoable, redoable, and exportable as versioned JSON.

03 / EDITOR

Agent IDE

Read every line. Decide every change.

The integrated Vim-style editor keeps code completion, diagnostics, definitions, run context, and agent changes in one inspectable workspace.

@@ -48,6 +48,11 @@ struct Factaccepted

4848pub dataset: RecordId,

49pub code_path: Option<String>,

50pub code_start_line: Option<usize>,

51pub code_end_line: Option<usize>,

52pub code_symbol: Option<String>,

4953}

@@ -211,6 +216,13 @@ impl GraphStorepending

211216self.ensure_dataset("code").await?;

217let fact = self.create_code_location(context).await?;

218if let Some(target) = selected_fact {

219 self.relate(&fact, "relates_to", target).await?;

220}

212221Ok(fact.id)

Real code intelligence

Rust-analyzer provides completion, definitions, and UTF-16-correct diagnostics inside the TUI.

Independent hunk review

Accept or reject each change without losing the agent’s original before-state.

Code becomes memory

Pin a selected symbol or range as a navigable graph unit and relate it to the current fact or run.

Conflict-safe undo

Before applying a decision, UIntellAgent verifies that disk state still matches the reviewed change.

04 / SAFETY

Execution boundary

Agency with explicit control surfaces.

The same permission engine sits between model intent and every tool. Dangerous actions wait for confirmation; autonomous runs remain bounded by checkpoints, policies, and review gates.

Intent

The model proposes a structured tool call with arguments visible in the timeline.

captured

Policy

Shared rules classify the tool, workspace path, command, and requested side effect.

evaluated

Sandbox

Code execution enters Bubblewrap by default, with unsandboxed mode requiring an explicit override.

isolated

Evidence

Exit codes, output, file versions, and verification results return to the durable run ledger.

recorded

05 / GUIDES

Engineering guides

Understand the runtime before you trust the run.

Original technical references explain AI agents, Rust orchestration, terminal coding, graph memory, checkpointed autonomy, and tool permissions from implementation details rather than marketing claims.

AI agent fundamentals

What is an AI agent?

Learn how AI agents combine models, tools, memory, permissions, and verification to complete multi-step work beyond a chatbot response.

9 min read

Runtime architecture

Building an AI agent in Rust

Explore how a Rust AI agent uses Rig, Tokio, Ratatui, typed tools, and explicit permission boundaries for reliable local execution.

10 min read

Terminal development

A terminal AI coding agent you can inspect

See how a terminal AI coding agent unifies chat, tools, code editing, graph memory, and durable runs without hiding file changes.

9 min read

Persistent knowledge

Graph memory for AI agents

Understand AI agent graph memory with SurrealDB: inspectable facts, typed relations, datasets, visual layout, safe queries, and repair history.

11 min read

Reliable autonomy

Durable AI agent runs

Learn how durable AI agent runs use atomic checkpoints, result ledgers, cancellation, resume, verification, review, and bounded repair loops.

10 min read

Execution safety

Tool permissions for AI agents

Design AI agent tool permissions with shared policies, explicit confirmations, path boundaries, Bubblewrap isolation, and authenticated gateways.

10 min read

06 / RUN

Run UIntellAgent

Bring the agent into your terminal.

From the project directory, choose a provider, verify the environment, and open the unified interface. SurrealDB starts automatically for the default local configuration.

Stable v1.0.2Rust 1.94+Linux x86-64local graph store

RUNTIME / INSTALLInstall UIntellAgent in one command

Downloads v1.0.2, verifies the pinned release checksum, and installs the binary to ~/.local/bin.

$curl --proto '=https' --tlsv1.2 -fsSL https://uintell.org/install.sh | sh