Skip to main content

Accountable agent-team runtime for complex work

Project description

Banksia recursive team mark

Banksia

Accountable AI teams for complex work.

Get started · Documentation · Workflow examples

Turn subagent workflows into accountable AI teams

Banksia is a no-code AI-team runtime for developers, researchers, and anyone who uses multiple agents to complete complex work accountably.

You operate Banksia as a product rather than embed it as an agent SDK; the controller owns published teams, run state, and completion.

Design reusable responsibilities without hard-coding how the work unfolds. Give the team one complete prompt, follow meaningful progress, respond when a real decision needs you, and receive one exact Result.

From ad-hoc subagents to a reusable team

Ad-hoc subagents Banksia
Recreate roles and prompts for every job Publish a reusable tree of named responsibilities
Reconstruct ownership from a transcript Follow controller-owned team, activity, and wait state
Treat provider completion as the outcome Accept only the lead's exact green or blocked Result
Recover by piecing together terminal sessions Recover from durable controller records; keep deliverables in ordinary workspace files

Banksia gives you:

  • explicit responsibility and delegation boundaries;
  • sequential, parallel, iterative, batch, or hybrid work chosen from current evidence;
  • independent implementation, research, criticism, and verification roles;
  • typed human decisions and managed command activity when explicitly granted; and
  • one inspectable Result with direct references to the files that carry the detailed work.

Get to a first Result

Banksia is not yet published to a package registry. Use a clean source checkout with Python 3.12 or newer, Make, and a supported Node.js/npm installation:

git clone https://github.com/ringlochid/banksia.git
cd banksia
make backend-install
make console-install
make console-package-assets
./.venv/bin/banksia init
./.venv/bin/banksia serve

make console-package-assets prepares the visual Console served by the source checkout. banksia init prepares local controller state, asks for one Task provider, and offers the separate Operator as an optional final choice. Rerun banksia setup later to change Task providers, Operator, or the default workspace.

Open http://127.0.0.1:18125/, go to Runs, and start a run with the production-feature-delivery Starter:

Add a guided configuration-import recovery experience across the API and Console. Preserve accepted imports, define the shared error contract, implement the service and user-facing behavior, verify the integrated recovery path, repair consequential findings, and return the release-readiness result with referenced files.

Watch who owns the work, read meaningful Activity, then read the exact Result and open its referenced files in your workspace. The getting-started guide includes the complete developer path and a researcher path using deep-research-and-decision-brief.

Start with a team

banksia init publishes eight provider-neutral Starters. Choose by the difficulty that makes one agent insufficient:

Starter Choose it for
production-feature-delivery A consequential feature crossing contracts, implementation boundaries, integrated verification, and release readiness.
incident-investigation-and-recovery An intermittent or high-impact failure needing competing hypotheses, supported recovery, independent verification, and prevention.
migration-and-modernisation A large migration needing inventory, dependency-aware batches, cutover, and stale-path removal.
deep-research-and-decision-brief A broad, consequential question needing independent evidence, claim verification, and one accountable recommendation.
decision-through-competing-prototypes A choice that should be tested through alternatives under one fair rubric instead of decided from prose alone.
idea-to-validated-demo A product idea that should become an evidence-backed position, working first demo, launch strategy, and credible pitch.
experiment-and-replication-program A substantial empirical or computational program needing explicit methods, durable execution, independent replication, and claim audit.
security-audit-and-hardening A security program needing attack-surface mapping, specialised audits, validated remediation, and adversarial re-verification.

These are not eight aliases for “implement and review.” Each team exists because parallel breadth, independent judgment, durable work, or evaluator-driven repair can materially improve the outcome. The complete Workflow catalog includes sample missions, expected deliverables, and guidance on when a simpler Workflow is better.

The product loop

Design → Publish → Run → Respond → Result

  1. Design a Workflow definition as a tree of responsibilities, not a timed sequence.
  2. Publish an immutable revision so every run has a stable team contract.
  3. Run that revision with one prompt, one workspace, and optional file references.
  4. Respond only when an explicitly capable Member opens a Human Request, or control the run when a current action is legal.
  5. Result is the lead's exact final green or blocked Checkpoint, with referenced files for detailed deliverables.

Current scope

  • Banksia runs as one loopback-bound process and is local-tool-first.
  • banksia serve is the foreground path. banksia service install selects a systemd user service on Linux or a current-user LaunchAgent on macOS.
  • Codex and Claude are managed providers. OpenClaw is a compatibility transport that you configure and operate.
  • A run uses one shared provider-visible workspace. Per-Member isolation and automatic merging are not current product behavior.
  • Human Request and Command Run capabilities deny by default. Revised Starters grant them only to the Members whose responsibility intrinsically needs a user decision or a long supervised process; capabilities never inherit.
  • File references record a workspace-relative path and optional description, not a snapshot. A referenced file can later change or disappear.
  • External MCP servers, reusable Skills, distributed delivery, and broad multi-user operation are deferred.
  • The Console supports the current authoring and operating paths and is currently desktop-oriented. Mature visual design and mobile/tablet experiences are deferred.
  • The native controller/runtime targets Linux and macOS. Linux is the currently evidenced source-development path; macOS release support still requires its native installed-wheel gate. Native Windows is deferred; WSL2 uses the Linux path.

Learn more

Banksia is open source under the MIT License, except for the visual Console in console/, which contains material derived from n8n and is distributed under the Sustainable Use License. That license limits use to internal business purposes or non-commercial or personal use, and permits redistribution only free of charge for non-commercial purposes. See console/NOTICE for attribution and the modification notice.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

banksia-0.1.0.tar.gz (734.0 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

banksia-0.1.0-py3-none-any.whl (912.4 kB view details)

Uploaded Python 3

File details

Details for the file banksia-0.1.0.tar.gz.

File metadata

  • Download URL: banksia-0.1.0.tar.gz
  • Upload date:
  • Size: 734.0 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.14.4

File hashes

Hashes for banksia-0.1.0.tar.gz
Algorithm Hash digest
SHA256 55f13e783aa2fdf2c8a39e54d14f8c79a4aefeb05fbf8f6322d3b5e8ffed2c2e
MD5 00df869aceae1785ab2591c360373470
BLAKE2b-256 ed5928ea03d90def5080cb5e9ae14bf41ba2a225fdb49db7aabecb8de3fc162c

See more details on using hashes here.

File details

Details for the file banksia-0.1.0-py3-none-any.whl.

File metadata

  • Download URL: banksia-0.1.0-py3-none-any.whl
  • Upload date:
  • Size: 912.4 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.14.4

File hashes

Hashes for banksia-0.1.0-py3-none-any.whl
Algorithm Hash digest
SHA256 c634d901e9a6a5fdcae764137ffae748d7c3d6c6a5876c10e2e7003a93648d3e
MD5 efad9fb9e857db4fcd73b1c51dbaff4c
BLAKE2b-256 965d663257fd15f4b4fee6888e3171fb4610f9d15394b9749b79d0ec6c036ed8

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page