Too much context for a small task?
Match information to the task. Bring in more when it helps, without making every step carry the whole history.
Relevant information, less repetitionAI solution building and optimization
Turn complex problems
into reusable solutions.
Break work into focused steps. Match each step with useful expertise, reusable tools and a suitable AI model, within the budget and permissions you set.
Designed to reduce repeated work and make better use of your token budget.
Private pilot · Invited access
Make the effort count
Is your AI spending effort in the wrong places?
These are the questions Baltor is designed to address at each step of the work.
An expensive model for every decision?
Compare suitable models and reusable code for each step, against your quality and spending requirements.
An approach that fits the workMissing the right domain expertise?
Bring relevant methods, examples and checks into the work, with sources and applicability kept visible.
Specific guidance, not generic guessesPaying to rewrite code that already exists?
Find an eligible implementation first. Generate new code for the gaps instead of reproducing a known solution as more output tokens.
Reuse what is already qualifiedAre the same mistakes showing up again?
Keep useful results and corrections. Compare proposed improvements on real tasks before making them the default.
Measured improvement, not blind repetitionOne workflow for the decisions behind the work
Spend more time on the problem.
Less time assembling an optimization stack.
Model selection, context sizing, tool choice and code reuse belong together. Baltor's goal is to let you configure, compare and improve those decisions against the same requirements.
You define success.
Set the quality standard, budget, allowed tools and data access. A cheaper approach is useful only when it still satisfies the task.
See the approach →Building blocks for better solutions
Put knowledge, code and experience to work.
Useful intelligence includes how to approach the work, code that can do it, what happened before, and what you want done differently.
Context Intelligence
Methods, questions, constraints, examples and output contracts that guide the work.
For the import task
A field-definition guide and a checklist for missing values.
Code Intelligence
Reusable functions, tools, packages and workflows with declared inputs, outputs and effects.
For the import task
A qualified normalizer with its dependency and verification records.
Runtime History and Solution Intelligence
Saved outcomes, failures, repairs, measurements and solution information that can inform a new task.
For the import task
A previous import's failed assumptions and the checks that detected them.
User Feedback Intelligence
Scoped corrections, priorities and instructions supplied by a person.
For the import task
“Keep uncertain matches for review. Never overwrite the original file.”
The four layers describe the broader product. The hosted pilot currently provides a small Context Intelligence example; it does not yet offer a populated catalogue across every layer.
A useful starting point for every step
Plan, build and review with the information each step needs.
Explore how a task becomes focused assignments, each with selected material and a way to request more when the work calls for it.
Explore an example taskFor planning
Relevant methods, requirements and previous findings
For building
Eligible code, selected skills and clear output requirements
For reviewing
The candidate result, acceptance criteria and known failure cases
Explore the first Baltor pilot
Start with reusable intelligence.
Invited users can search permitted material, inspect its source and download a selected revision. The broader solution-building workflow, public registration, paid subscriptions and complete native harness onboarding are still in progress.
Token savings depend on the task and configuration. No percentage reduction, replacement of specialist teams or guaranteed daily performance gain is claimed for this pilot.
Your workspace
Welcome back.
Connect with the service access token supplied for your private pilot account.
Your model provider key is different. Keep it in your local harness, not in this form.
Don't have access? Check registration status.Sign in with email
Sign in to the pilot
Not connectedEmail sign-in is not enabled yet. This form checks real service access; it does not create an account or a subscription.
Registration
Public sign-up is not open yet.
The first release is an invited pilot. If you already have a service access token, you can sign in and use the hosted workspace.
What remains before registration opens
Verified email delivery, customer account provisioning and the complete sign-up journey must pass their integration checks.
Create your Baltor account.
Confirm your email before using your account. Creating an account does not start a subscription or give access to your local files.
Intelligence workspace
Find material for your next task.
Inspect an exact reference before fetching its body. Your harness runs in your environment.
Selected material
Search intelligence
Search returns references. Fetching a body requires a separate access check and may record usage.
Start with an assignment
Connect, describe the material you need, then inspect the returned source and permissions before downloading.
Your account
Access and usage.
Current service identity
Not connectedSign in to see your service identity, usage and available subscription settings.
Recorded usage
Service usage is separate from the model calls your local harness makes.
Connect to inspect your permitted usage records.
Subscription access
Only configured plans are shown. A completed checkout does not itself confirm that access is active.
Connect to check this service's billing configuration.
Administrator access
Manage test-access tokens.
Sign in with an administrator service token. Email is not required.
Administrator sign-inCreate a test token
Tokens share the selected tenant's material and usage. A token is not a separate private account.
Save this token now
Shown once. It cannot be recovered from the service. Store it in your secret manager.
Issued test tokens
Revocation takes effect on the next service request. It does not recall material already downloaded. Bootstrap administrator keys are managed separately.
How it works
Big problems.
Clear steps.
Start with what you want to achieve. Break the work into steps, give each step relevant expertise and tools, then check the result before moving on.
The plan can change as you learn more. A step can ask for missing information, try another approach or send unfinished work back for improvement, within the limits you set.
See what each step needs
Prepare a customer import without changing the original.
Choose a step to see its information, tools and expected result. The person inspecting the data needs different material from the person checking the finished import.
How this step could run
Information for this step
Selected filesWhat it can do
What it cannot access
What you get
When this step is finished
This is an illustration, not a recorded customer result. Selecting a step does not access your files or call a model.
Useful information, when it is needed
Give each step a useful starting point.
- FindLook for relevant guidance and reusable code
- ChooseCheck that it fits the task and your permissions
- PrepareGive the step its selected files and instructions
- WorkUse an appropriate model, tool or existing code
- CheckTest the result against your requirements
A useful briefing is not always the shortest one. Start with relevant information and add more when the work reveals a gap.
What if a step needs more information?
It can search for additional material or ask a question. Your permissions and remaining budget still apply. New instructions cannot grant access to private files or allow spending by themselves.
If an action times out, first check whether it happened. Trying again must not send a second email or repeat another change by mistake.
Use what suits the work
Not every step needs a large model.
The goal and quality checks stay the same. The way a step does the work can vary.
A focused AI workspace
Give your chosen development tool a clear brief, selected files and the permissions needed to build or repair the missing pieces.
A smaller, focused request
A decision may need one model request rather than a complete development session. Use a compatible model service you have already configured.
Run code that already fits
Use reviewed code when it meets the requirements. That step may need no model call and no newly generated code.
Models may run on your machine or through a provider you choose. You supply access; Baltor does not need to install or host the models for you.
Your tools, with help from Baltor
Get useful material. Keep control of your work.
Available in the private pilot
Invited users can sign in with a service token, search the small example library, download permitted files and view usage. Administrators can create and revoke test tokens.
The complete example above is still being tested from start to finish. Public email signup, paid subscriptions and automatic setup of every step are not open yet. We have not established token savings or overnight task completion for this release.
Client setup · Private pilot
Give your tools access
to useful intelligence.
Use your existing client and a scoped service token. You do not need to install or host a model to connect.
Choose a connection recipe.
Loading documented compatibility information.
Configuration
Merge this entry into your existing configuration. Do not replace your other settings.
Loading the secret-free configuration.
Source: Client documentation
Keep the token out of your files.
Set BALTOR_SERVICE_TOKEN through your local secret manager or terminal environment, then start the client from that environment. Use the test token issued for your tenant, never an administrator token or a model provider key.
Temporary setup in Bash
This prompt hides the value while you enter it. The value is not part of the command history. The client process inherits the environment variable.
read -rsp "Baltor service token: " BALTOR_SERVICE_TOKEN export BALTOR_SERVICE_TOKEN
After closing the client, clear the terminal variable with unset BALTOR_SERVICE_TOKEN. Do not run environment dumps or record the terminal while handling secrets.
Your client may need a restart after an environment change. Its own model access remains a separate connection.
Check the service, then the client.
This browser check initializes the protocol and lists tools using your current service connection. It does not run an agent, retrieve file bodies, or spend model credits.
Not tested. No model calls are made by this check.
Then check your installed client
Choose a client above.
Configuration, connection, retrieval, native loading and useful task completion are separate checks. A green connection does not qualify all five.
Try one useful retrieval.
Inspect a permitted reference and verify the selected file.
First-use example
Inspect first.
Fetch only what you need.
Practice the real retrieval path with the pilot's small input-review procedure. No model call is needed for this browser exercise.
1. Search permitted material
Sign in with your pilot token, then search for review inputs. The query returns metadata and exact references, not file bodies.
This prepares a query in the workspace. You choose when to search.
2. Inspect the result
Open Source, integrity and access. Check the identity, source, license, digest and qualification basis. The current example is host-attested diagnostic material, not independently qualified commercial intelligence.
Access is tenant-specific. An empty result can mean your account lacks the grant; it does not mean the item is available to everyone.
3. Fetch the exact revision
Select Fetch exact revision only if body access is permitted. The service checks your current grant again. The browser verifies the downloaded bytes against the selected digest before saving them. A download can record usage.
A downloaded file is still material to inspect. This page does not execute it, install a plugin or confirm that your client loaded it.
4. Inspect what happened
Open your account and refresh usage. Keep the exact reference with any later task result. If a download times out, retry the same selection in the same page session to reconcile its request identity; do not assume the first request did nothing.
What this example proves
You can exercise tenant-scoped search, deliberate selection and integrity-checked delivery. It does not prove model quality, lower token cost, a complete native harness task or automatic improvement.
Pilot access and data
Know what you share.
Control what can run.
Connecting to the intelligence service grants access to permitted material. It does not give the service permission to run commands on your computer.
Service access and model access are separate
A service token identifies a scoped tenant connection. The service stores token digests, checks scope and expiry, and can revoke access. Your model provider keys belong with your local client or approved credential broker. Do not enter them into the website.
The browser holds its service token in page memory only. Closing or reloading the page clears that connection. Avoid shared devices, untrusted extensions and screenshots of credentials.
Selected files do not grant execution authority
Search returns permitted metadata. Fetching a body rechecks access and its exact identity. A digest proves that bytes match the selected reference; it does not prove that the material is safe, correct or useful.
Inspect unfamiliar code and plugins. Use a confined workspace and a sandbox appropriate to the work, with explicit file, command and network permissions. A shared container does not isolate one process from every other process inside it.
What reaches this service
Search text, requested references and service authentication reach the server when you use the workspace. The service records access and usage metadata. The website does not automatically upload your project files, model keys or complete execution traces.
Do not send private customer content during the pilot. The final retention policy, deletion workflow and consent controls remain launch work. Embeddings and derived features must not be treated as anonymous data.
Current operational limits
The pilot runs on one Fly machine with an attached persistent volume. A local backup-and-restore exercise has passed. This is not a multi-region service, an availability guarantee, an independent security audit or a compliance certification.
Public registration and live subscription charging are not open. Pilot access comes from your operator, who can issue and revoke test tokens. A test token cannot delegate administration.
Before sensitive or consequential work
Agree on the data policy and task permissions with your operator. Keep provider credentials out of prompts and retrieved files. Do not repeat an external action after an uncertain outcome until its state has been reconciled.
Setup guide
Connect a client.
Keep your keys private.
There are two different connections: access to an intelligence service, and permission for your harness to call a model provider.
Technical runtime reference
The public website calls a unit of work a step. The implementation uses the canonical Loop runtime described here and in the repository.
Runtime classification and role profiles
Complete behavior of a discrete cognitive or act step Loop node
A discrete cognitive or act step Loop node is an independently governed instance of the Loop runtime responsible for one clearly defined cognitive step or action. A cognitive step might interpret information, identify a missing requirement, compare alternatives, or evaluate a result. An action might inspect a directory, build software, execute a test, create an artifact, or send an authorized email.
Each discrete cognitive or act step Loop node receives the context, instructions, skills, plugins, tools, and working files relevant to its assignment. Essential information can be supplied directly, while additional information can remain in centralized storage behind authorized, versioned references. It does not automatically need the entire task history or every available tool.
A separately initialized harness process, such as OpenCode, Pi, Codex, or a custom implementation, can perform the assignment. When explicitly permitted, another harness can attempt the same assignment after a failure. The assignment's contracts, permissions, history, and remaining authority persist across those attempts.
Discrete describes the scope of the assignment, not a restriction to one attempt, one model call, or one output. A discrete cognitive or act step Loop node can examine whether an observation matches its expectations, identify a problem, repair or change its approach, and repeat until its declared completion conditions are satisfied.
Alternatively, a discrete cognitive or act step Loop node can publish an initial candidate output and continue working while its continuation conditions and authority permit. It can produce additional alternatives over time, including alternatives that are better, worse, or useful under different circumstances. Consumers must identify exactly which output they used. Publishing an output does not necessarily mean that the producing assignment has finished.
For externally consequential actions, continued operation does not authorize repeated effects. For example, generating alternative email drafts can continue, but sending an email requires its own authorization and protection against duplicate delivery.
Harness Intelligence is a provisioning view over the four persistent layers. It is not a fifth persistent layer.
Runtime Memory is the temporary note board for one run. Saving a note does not promote it into reusable intelligence.
1. Install the local runtime
Use the release and installation instructions supplied by your service operator. When installing this repository from an approved checkout, the serving dependencies are optional:
python -m pip install '.[serving]' loop-engine service --help
The intelligence service can run locally or remotely. A remote model call still sends the selected context to that provider.
2. Obtain service access
Your operator creates a tenant and binds its access before issuing a scoped key, or binds your identity-provider subject. There is no automatic account creation from an unverified email address or a token's tenant claim.
For Model Context Protocol clients, configure the endpoint shown in the workspace and supply its service token through your client's supported secret mechanism. Use the exact protocol version returned by the service. Do not put tokens in a URL, a shared Markdown file, or a checked-in client configuration.
3. Bring your own model keys
Keep model credentials on the machine or broker that performs the model call. Store only a secret reference in shared configuration. The browser workspace never asks for a model provider key.
credential_ref: env:YOUR_PROVIDER_KEY allow_network: false
This is a configuration pattern, not an executable universal provider file. Use the selected adapter's documented schema. Enable network calls only after choosing the exact provider, model, task allowance, and disclosure scope.
A native harness may use its own supported sign-in. A brokered tool can instead use a host-held credential on behalf of a scoped assignment. Neither arrangement removes the provider's terms, account limits, or required permissions.
4. Verify the connection
Connect to the workspace and confirm the tenant, namespace and scopes. Search for a permitted item. Inspect its identity, qualification basis and digest. Fetching a file does not prove that your harness loaded it or used it correctly.
If access expires, reconnect with a valid token. If a download or checkout has an uncertain outcome, inspect its recorded state before repeating the operation. Do not assume a timed-out action was cancelled.
Privacy and telemetry
Do not include keys in prompts, intelligence bodies, screenshots, logs, or feedback. Embeddings and raw context are confidential data, not anonymous telemetry. Choose retention and improvement consent separately from execution access.