PaloSpring
We are aware that the agentic IDE world is already well saturated. PaloSpring Autonomous Multi-Agent Development Architecture attempts to chart a different trajectory.
The mission is not faster coding. It is continuity. AMDA treats development as an operational process where intent, rationale, and decisions are preserved as artifacts across research, architecture, implementation, and validation.
PaloSpring is also the first operational environment for the memory problem Palo Bloom is intended to address.
[Wide hero: saturated landscape of agent tools fading into the background, a single structured pipeline path emerging forward with labeled stage artifacts flowing between gates]
Open source. Self hosted. Structured autonomy.
Local control first
Self hosted runtime. Stage gates. Council review at round close.
[Live terminal demo beside copy: pipeline stages as a vertical gate sequence, each gate showing pass/fail state and the artifact handed to the next stage]
Type "palospring".
Choose a workspace. Write the directive. Submit to begin the run.
[Screen recording: workspace picker, then PALO::TASK panel with a directive being submitted and the first pipeline stage activating]
Research before implementation.
Prior artifacts are reused when possible. Otherwise pre research runs. The research brief must clear its gate.
[Diagram: workspace state on the left, evidence queries in the center, a confirmed research brief artifact on the right with a locked implementation stage behind a gate]
Architecture as artifact.
Blueprint from research brief. Preflight checks. Critique before execution. Each stage receives the prior artifact, not a conversation backlog.
[Blueprint document expanding into module tree and interface map, with preflight and critique gates shown as two checkpoints before code]
Implementation and validation.
Code from blueprint. Integrate. Run the validation ladder. QA closes the round.
[Split view: code branches merging through an integrator, then a validation ladder with test results and a green QA pass state]
The Council
Deliberation is separate from execution. Issues route to the Council. Votes are structured. Overrides are logged.
[Council panel in the terminal: agent names, GRANT/DENY/ABSTAIN votes, confidence bars, and an audit log entry written to .palo/audit/ on override]
Stage routing
Different models per stage. Local or remote. Optional.
[Routing diagram: research, architecture, and implementation stages each pointing to a different model provider, with health scoring rerouting around a failed node]
Open Dataset
Opt in research data
Anonymised task and output pairs. Off by default. Public dataset.
[Settings toggle for dataset contribution, with anonymised run artifacts flowing into a public research corpus]
Effort Mode
Low to Extra High. Controls research depth, critique rounds, and review before Council.
[Four step slider with increasing pipeline depth shown as more research loops, critique passes, and council review rounds]
Start here
Evaluate it yourself.
Open source. Self hosted. Documentation and repository available now.
[Closing frame: PaloSpring pipeline on local hardware connected to Palo Bloom memory infrastructure, with continuity preserved across runs]