Saga Agent / Intent into action

Put AI to work on your terms.

Own the work context. Bring your model keys. Decide what the agent may do. Saga Agent is an AI harness that connects context, models, tools and permissions to work you can review—in VS Code or the CLI.

Explore the features See a task walkthrough

Your context. Your models. Your control.

01

Own the work context

Keep your files, work records and reviewed memory connected in your chosen environment. Decide what context to share with configured models and tools; remote inference sends selected context to that provider.

02

Bring your own model keys

Choose supported frontier or open-model routes for the task, budget and data policy. AI is BYOK: provider charges are yours to manage, with no bundled token allowance implied.

03

Keep authority over action

Set scope, handle approvals and inspect changes and checks. A model suggestion becomes action only through the configured runtime controls.

Explore. Work. Plan.

Choose how you want to engage: understand a problem, pursue a defined goal, or organize dependent work. The mode sets the workflow; you keep direction.

01

Explore the context

Ask questions, investigate files and trace how a system works. Move from conversation to proposed edits when you are ready to review them.

02

Work toward a goal

Define the objective and what done means. Let the agent work through the steps, inspect results and run checks within the scope you authorize.

03

Plan the execution

Organize a complex task into execution steps. Review the intended scope and sequence before letting the agent act.

More of the work. More of the record.

Search, edit, verify and continue substantial tasks through the package supplied by your network.

01

Find the context that matters

Search files and inspect code structure before acting. Focus the model on the relevant symbols, references and diagnostics so you can spend less time assembling context.

02

Make code-aware changes

Use syntax-aware outlines and symbol edits for work across files. Inspect proposed changes, accept or reject them, and use the supported undo path when needed.

03

Set boundaries for action

Review requests for additional paths or commands before a task expands. Handle pending approvals in the workbench.

04

Check the result

Run authorized tests and diagnostics, inspect diffs and examine Git state. Keep failures visible while the agent works through the next correction.

05

Keep work moving

Queue established goals for later work and return to pending tasks or approvals. Reviewed memory notes can preserve useful context across sessions.

06

Retain evidence for review

Keep work records, changed paths and checks connected. Inspect the resulting evidence capsule to understand what ran and what it established; acceptance remains a separate review.

Example: update a shared validation rule.

Illustrative task, not a recorded run: change a shared rule and identify the tests needed to review it.

A request becomes a model-proposed step, checked action, observation, updated work record and deliverable for review.
  1. Define a goal, context and constraints.
  2. Let a configured model propose the next step.
  3. Check permissions before using tools.
  4. Observe the result and update context and the work record.
  5. Deliver the artifact, checks and remaining gaps for review.
Conceptual harness workflow adapted from the technical overview; the task below is illustrative.
  1. 01

    Scope: locate the validator, its callers and dependent tests; limit changes to the agreed files.

  2. 02

    Expected work: propose the rule change and test updates, then run the authorized checks and retain their results.

  3. 03

    Review material: a diff, test results and any unresolved gaps, ready for a person to assess against the requested change.

Choose models. Connect tools.

Use the VS Code panel or command-line runtime with supported model routes and opt-in tool connections.

Saga Agent connects configured frontier-provider or local/private model routes with context, tools, permissions and work records.
  1. Choose a supported frontier or open/private model route.
  2. Bring your own keys and select which context is shared.
  3. Use the harness to manage tools, permissions and work records.
Conceptual model-choice diagram adapted from the technology overview. Routes and tool behavior depend on the offered package and configuration. BYOK alone does not guarantee data retention or privacy settings.
  1. 01

    VS Code: interact with the agent and inspect edits, tasks and approvals in the panel.

  2. 02

    CLI: run authorized tasks and inspect their status and evidence from the terminal.

  3. 03

    Usage: follow token and cost records for a run.

  4. 04

    MCP: configure trusted tool servers and their permissions before using them.

Connect execution to GAME Plan.

Optional integration preview: take an authorized task from GAME Plan and return the work record and checks for review. GAME Plan coordinates the wider work; Saga Agent carries out the bounded task.

Explore work and attribution planning

Start through your network.

Available to SagaHalla builders and Aregis members through scoped access; retail offerings and pricing are still being finalized. GitLab access is coming soon and requires the qualifying subscription, signed applicable agreement and verified network/product authorization.

01

SagaHalla Network

Our home network for builders, early adoption and contributions to the tools. Bring a task or integration you want to explore.

Join SagaHalla with Agent interest
02

Aregis Network

The first external licensed network, for scoped engagements and regulated workflows under its applicable terms.

Join through Aregis
Read the product and setup overviewReview pricing and accessReview source licensing