Start with one task. Your coding agent uses Weft’s included workflow, checks, and project memory to launch and manage workers. You have one conversation to follow, with less to keep in your head. Reuse the process, then adapt it as you go.
Free and open source. Local workflows and memory included. No subscription or Cloud service required.
ONE REQUEST / MORE WORK MOVING Example
You“Add an export.”
✳
Your agentUses Weft’s tools
Working
Saved workflowPlan→Build→Check
⌘APIBuilding
>_CLIBuilding
✓ChecksRunning
↗
Three workstreams. One request.
Your agent delegates independent work and keeps the results together.
▤
PROJECT MEMORYDecisions + the reasons behind themSaved in your local project
✓
EXAMPLE / AGENT ASKS YOU
Keep the original timestamps?
EXAMPLE / HUMAN REPLY
“Keep them. The dates matter.”
↳
A fresh agent session · same project“Same checks. Keep the original dates.”
01 / 06
Read all six steps
Three workstreams. One request.Your agent delegates independent work and keeps the results together.
A finding. A revision.Your tests find a missing field. Revise, then check again.
Your judgment belongs here.No prior answer about dates. The agent brings the question to you.
Your answer stays with the work.Your agent records the choice and its reason in project memory.
A new task. A known way to work.A fresh agent opens the saved workflow for a new task.
The next task starts prepared.The next task reuses your workflow, checks, and decision.
Illustrative workflow · Local records + your checks. Scripted example · plays through and repeats. Your agent runs parallel work, checks results, asks for a missing decision, and records your answer. A later task reuses the workflow, checks, and decision. Read the six steps for the complete example.
Preview · measurement is disabled for this view.
START WITH ONE CHANGE
A working process from the first task.
Bring a small change and the checks you already use. Your agent can run it through the included workflow, record the result, and reuse the process for the next task.
Have your agent check prior decisions, ask when the answer is missing, and record your reply.
YOUR PROCESS / A REUSABLE ENGINE
AI writable. Human readable.
Ask your agent to write the workflow YAML. Read the stages, choose the checks, and save the process with your code. Weft runs it again for the next task.
Start with the included software development flow. Write your own plugin when you need new tools or checks.
The same Weft engine.Your stages. Your checks. A reusable process.
Illustrative flow · customize stages and checks in your project.
Run the next task with your revised process and the same engine.
Start with the included software development flow.
Add a review stage to your workflow definition.
Run the next task with your revised process and the same engine.
PARALLEL WORK / REQUIRED CHECKS
Run tasks in parallel. Keep your checks in place.
Your agents can work on separate tasks at the same time. A failed check stops the affected task from moving on, so the agent can fix it before continuing.
AGENT / WORK RECORDOSS · local workflows + your checks
You
Add search and fix the CSV export.
Your agent → Weft
Record two independent tasks and their acceptance criteria. Dispatch a local worker for each.
Worker agents
One agent builds search while another fixes the export.
Required checks
Search passes. The export test still fails: some rows are missing. Hold that task.
Export agent
Fix the missing rows and rerun the required checks.
Work record
Export checks pass. Both results are ready for you to inspect.
Illustrative OSS workflow: two tasks, each with required checks. Playback changes this example only.Source and scope
This is an illustrative example using local OSS tasks, managed workers and configured workflow checks. Your repository supplies the tests. Each task follows its own workflow; a failing required check holds its transition. Historical parallel-work examples are on the examples page.
EVERYTHING RUNS LOCALLY
Your machine. Your agents. Your process.
Project memory, tasks, workers, workflows, and gates are included in OSS. State stays in a local SQLite database inside your project. Free and open source.
Working across machines?
The optional commercial Cloud plugin adds hosted shared state and AI review. OSS keeps working independently.
Start with one task. Your coding agent uses Weft’s included workflow, checks, and project memory to launch and manage workers. You have one conversation to follow, with less to keep in your head. Reuse the process, then adapt it as you go.
One shared process: tasks · designs · decisions · evidence Process illustration. Your workflow defines the actual stages and checks.
Preview · measurement is disabled for this view.
A PROCESS YOUR AGENTS CAN USE
A working process from the first task.
Bring a small change and the checks you already use. Your agent can run it through the included workflow, record the result, and reuse the process for the next task.
Your agent delegates parallel tasks. Local workers keep changes in separate worktrees.
A memory for the project
Keep decisions and their reasons in a local knowledge base. A fresh agent session can retrieve them before continuing the work.
Accept work with evidence
Define stages, enforce the checks that matter, and keep human checkpoints visible when a decision needs your judgment.
BUILDING WEFT WITH WEFT
Three workstreams. One acceptance process.
Independent tasks can keep moving while a dependent step waits for its check. Here is that process from your agent’s point of view.
AGENT / WORK RECORDOSS · local tasks + your agent
You
Move the migration forward. Keep independent work in parallel.
Your agent → Weft
Record three tasks, their acceptance criteria and dependencies in the local project.
Your agent
Delegate separate workstreams to agents in separate worktrees.
Worker agents → Weft
Two changes complete. The third task confirms no change is needed. Results recorded.
Repository check
Remaining consumers found. Hold the dependent removal step.
Your agent
Assign the correction. Recheck consumers before continuing.
Illustrative OSS replay, adapted from recorded parallel work. Playback changes this example only.Source and scope
The recorded June 2 migration launched three agents within 26.6 seconds and received their completion responses in about 40 minutes. One lane was a verified no-op. A later correction migrated 70 consumers; the final sweep found and migrated seven more. Claude launched the agents and applied the consumer check. This replay translates that practice to local OSS records; it is not a recording of native worker dispatch.
Inside the recorded migration
June 2, 2026: three agents launched within 26.6 seconds and received completion responses in about 40 minutes. Two lanes changed code; one verified that no change was needed. A later correction migrated 70 consumers, and a final sweep found and migrated seven more before the check cleared.
The historical work used Claude Code and Weft with the Cloud plugin. The replay adapts the coordination pattern to OSS. Your agent delegates; OSS keeps local tasks, dependencies, decisions and workflow evidence.
HOW IT WORKS
Ask your agent. It uses Weft.
Tell your agent what you want to build. It uses Weft to organize tasks, consult project memory, and coordinate the agents it delegates to. Managed local workers run the configured workflow in separate worktrees.
YouSet the outcome and rules
Your agentPlan and dispatch the work
Worker agents + WeftRun stages and gates
YouDecide where needed
What changes in my day-to-day workflow?
You continue talking to your coding agent. Ask it to record work, keep the relevant context, and dispatch tasks through Weft. A managed local worker executes the active task workflow. You can inspect state and evidence through your agent or the CLI, and review or merge results through your normal repository process.
Weft provides local agent coordination today. Unified execution placement and design-to-implementation session continuity remain planned.
YOUR JUDGMENT, KEPT WITH THE WORK
Ask once. Keep the answer.
Tell your agent to check prior guidance before a significant change. When the answer is missing, it brings you the choice. It records your answer and its reason so a later agent can retrieve them.
AGENT / WORK RECORDOSS · decisions in your local project
Your agent
The importer could change existing dates. That changes the meaning of the records.
Your agent → Weft
Search prior decisions about import timestamps.
Local decision records
No matching answer found.
Your agent → you
Keep the original dates, use the import date, or ask before changing them?
You
Keep the original timestamps.
Your agent → Weft
Record the decision and its reason in the local project.
Keep the original timestamps.Reason: People use dates to understand the order of their records.
A later agent session
Retrieve the decision before continuing the importer.
Keep the original timestamps.
Illustrative replay, based on recorded agent workflows. Playback changes this example only.Source and scope
Your agent follows the search → ask → record → retrieve policy using Weft's tools. A recorded June 18 episode includes an empty search, three options, a human answer and a saved decision. Related decisions were retrieved and applied by later workers. The importer here is an illustrative example, not a literal transcript. The decision stays in the local project; sharing it across separate machines is a Cloud capability.
A WORKED EXAMPLE
From one request to work you can accept
Your repository’s tests give an implementing agent a concrete failure to address. Configure verification as a required gate when the work must wait for those checks.
AGENT / WORK RECORDOSS · your repository's build and tests
Implementing agent
Importer implemented. Run the required build and test verification.
Repository test
FAIL · importing an entry changes its original timestamp.
Workflow
Verification has not passed. This transition is held.
Implementing agent
Preserve the source timestamp. Add a regression test.
Repository checks
PASS · build and tests. Verification evidence recorded.
Your agent
The required checks passed. Inspect the result before merging.
Illustrative configured local verification workflow. Playback changes this example only.Source and scope
OSS supplies local workflow stages, structural checks and build/test verification. The test shown is an example of a repository test you supply. This is not hosted AI review, and a passing test does not replace your acceptance judgment.
YOUR MODELS, YOUR ACCEPTANCE PROCESS
A smaller engine for a bounded job.
Your agent can give a well-defined implementation to a lower-cost model and keep a stronger supervisor on the result. The task and its acceptance criteria give both agents a common reference.
AGENT / WORK RECORDYour agent · your model accounts
Your agent
Small, bounded task: remove a duplicate reachability check.
Weft work record
Task, acceptance criteria and the allowed files are available to both agents.
Pi · DeepSeek Flash
Implement the functional change.
Claude Opus · supervisor
Check the premise in the source. Demonstrate the regression check fails before the fix and passes afterward.
Supervisor
Run the full repository checks. Inspect the patch.
Work record
Verified change merged as PR #1195.
Adapted from a recorded implementation and supervision workflow. Playback changes this example only.Source and scope
One recorded Pi implementation turn took 26.905 seconds, with a model-reported cost estimate of $0.008538006. That excludes supervision, review, tests and CI. The supervising agent selected the models; this is not automatic model-to-complexity classification by Weft. This pattern uses your own coding engines and model accounts, with no Cloud requirement.
CONFIDENCE AND CONTROL
Decide what must pass.
Put acceptance criteria in the process. Use build and test verification, bundled structural checks, and executable gates you supply. When an enforced gate fails, the transition stays held until its conditions are met.
What checks ship with OSS, and what can I add?
The bundled checks cover structural requirements and selected static patterns, with build and test verification as a separate check. You can add executable gates through plugins, or use the confined script runner through its direct gate interface. A name in YAML alone is not an executable check. Checks establish only what they actually measure.
Choose a workflow and bind gates at the transitions you need to control. An enforced binding makes its passing verdict a condition of advancement. Document the inputs, failure conditions, and evidence for any gate you supply.
AI-powered review can be added by a review provider. It is not a guarantee supplied by the bundled deterministic checks.
FREE AND OPEN SOURCE
Your agents, working together on your machine.
Coordinate your agents with tasks, decisions, workflows, and checks. State stays in a local SQLite database inside your project.
Free and open source under Apache 2.0. No subscription or Weft Cloud account required. Bring your own coding agent and model access.
Add shared state or AI review.
Hosted state lets people and agents coordinate across machines. Cloud reviewers examine plans, designs, and code, giving agents findings to address before work advances. Useful for teams—or for a solo developer who wants AI review.
The commercial Cloud plugin is optional. Local tasks, decisions, workers, and gates are included in OSS.
For developers already using coding agents who want to delegate more work with confidence and control. Start with one project and the checks you already use. Add parallel workers and custom gates as your process grows.
Explore setup before you commit
Once public installation is available, install the Weft binary and register it with your MCP client. Guided setup supports Claude Code, Cursor, and Codex. Other agents can use Weft through their MCP integration or the CLI.
# After public launch and installation:
weft install --client claude-code
# or:
weft install --client codex
weft init
Restart your assistant, ask it to create a Weft project and a small task, and start with checks you already trust. Your agent remains the interface; you can use the CLI directly whenever useful.
OSS launch is being prepared. Public installation and repository access will be announced here. You can explore the process now and ask for launch updates.
BEFORE YOU START
A few practical questions.
Will OSS remain free?
Free and open source under Apache 2.0. No subscription or Weft Cloud account required. Bring your own coding agent and model access.
Why would I add Weft Cloud?
Hosted state lets people and agents coordinate across machines. Cloud reviewers examine plans, designs, and code, giving agents findings to address before work advances. Useful for teams—or for a solo developer who wants AI review. The commercial Cloud plugin is optional. OSS keeps working locally.
Is this another coding agent?
No. You keep talking to your agent, such as Claude Code, Codex, or another compatible client. It uses Weft’s tools to manage work. Dispatched agents use the same system. The CLI is also available when you want direct control.
Does OSS require an account?
No Weft account is required. The bundled tools and SQLite storage work locally without a hosted service. Your coding agent may still need its own model access and network connection.
Can I choose different models for different work?
Your agent can choose the agents and models it delegates to. Weft gives them shared tasks, decisions, and a process. Managed workflows use their configured execution engine; automatic per-stage model selection is still planned. Model access and charges come from your chosen providers.
Does it guarantee correct software?
No. Stages and gates make the acceptance process explicit. Bundled review is deterministic and structural; it does not establish general semantic correctness. You choose meaningful checks and retain judgment over acceptance.
Can I define my own process?
Yes. Start with the included SDLC workflow and edit its declarative YAML definition to change stages, transitions, and gate bindings. Weft is a generic workflow engine; the bundled plugin supplies the software development tools and flow. Executable checks come from installed gate providers. Bind a gate as enforced where progress must depend on its verdict.
Can I use multiple agents?
Managed local workers can execute dispatched task workflows concurrently in separate worktrees. Unified placement across every execution path and design-to-implementation session continuity are still planned.
How will I get started?
At public launch, install Weft, connect it to your MCP client, initialize a project, and ask your agent to create a task. Guided registration supports Claude Code, Cursor, and Codex. Other clients can use their own MCP configuration or invoke the CLI. The repository is still private; join the launch list for availability.
Does it have a memory?
Yes. The local knowledge base records project decisions and their rationale. Agents can search it across sessions, including when you switch agents or models. Ask your agent to consult it before planning related changes.
I already built a software factory. Why use Weft?
Keep your agents, skills, and repository checks. Try one task through Weft before adopting more. It supplies the workflow engine, local work records, and managed workers, so you can change your process without writing all of that machinery yourself. Your custom checks and integrations still need maintaining.
Can I use it through Dots or another assistant?
If your everyday assistant delegates coding work to another agent, connect Weft to that coding agent. Direct access needs an environment where the agent can run the Weft CLI or connect to its MCP tools. Guided setup covers Claude Code, Codex and Cursor; a direct Dots integration has not been tested.
Why use a workflow if my agent can figure it out?
Your agent still figures out the change. A saved workflow gives it stages and checks to reuse; project memory carries earlier decisions. Inspect task history and findings to decide what to improve. Start with one task and the checks you already trust.
WHY THIS APPROACH
Structure makes delegation practical.
Weft grew from hands-on work with coding agents and research into how they fail. The aim is a process whose checks and evidence you can inspect, rather than relying on a confident completion message.
Get an email when Weft OSS is available. Start with one project, your existing coding agent, and checks you already trust.
Privacy and aggregate measurement
This site measures aggregate experiment exposures, section exploration, and successful new signups. Analytics sends only a fixed experiment name, a structure/copy category, and an event category. It uses no analytics cookies, persistent visitor identifiers, fingerprinting, or cross-site tracking. The six possible structure/copy choices can travel in links within this site. These shared categories identify no person. A new visit without that link context can show a different version. Hosting and network providers process requests separately. Signup emails are stored separately so we can send launch updates. They are not included in analytics. Aggregate counters are marked for expiry after 90 days. For privacy or deletion requests, contact the maintainer through michael.roth.rocks.
The aggregate report counts page exposures and events, not unique people. A failed signup is never counted as a successful registration.
Scenario-page entries use a separate experiment comparing a focused page with the homepage within fixed campaign categories. These categories can travel in links; they identify no person.