fabri 0.1.0__tar.gz

fabri-0.1.0/LICENSE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or Derivative
+          Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2026 Rushikesh Patade
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

fabri-0.1.0/PKG-INFO

@@ -0,0 +1,243 @@
+Metadata-Version: 2.4
+Name: fabri
+Version: 0.1.0
+Summary: A local memory + orchestration framework for custom LLM agents: tactical/strategic memory over execution traces, polyglot subprocess tools, agent-as-tool composition, and token-efficient TOON I/O.
+Author-email: Rushikesh Patade <pataderushikesh@gmail.com>
+License-Expression: Apache-2.0
+Project-URL: Homepage, https://github.com/Rushour0/fabri
+Project-URL: Repository, https://github.com/Rushour0/fabri
+Project-URL: Issues, https://github.com/Rushour0/fabri/issues
+Keywords: llm,agent,agents,memory,anthropic,claude,openai,qdrant,orchestration,toon,react,tool-use
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Developers
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Operating System :: OS Independent
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Topic :: Software Development :: Libraries :: Application Frameworks
+Requires-Python: >=3.11
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: qdrant-client>=1.9
+Requires-Dist: sentence-transformers>=3.0
+Requires-Dist: anthropic>=0.40
+Requires-Dist: tiktoken>=0.7
+Requires-Dist: pyyaml>=6.0
+Provides-Extra: openai
+Requires-Dist: openai>=1.0; extra == "openai"
+Provides-Extra: dev
+Requires-Dist: pytest>=8.0; extra == "dev"
+Requires-Dist: build>=1.0; extra == "dev"
+Requires-Dist: twine>=5.0; extra == "dev"
+Dynamic: license-file
+
+# fabri
+
+A local, open-source memory + orchestration layer for a custom LLM agent.
+Built around two ideas:
+
+- **SCOPE** (arXiv:2512.15374): treat the agent's prompt context as something
+  that evolves automatically from execution traces, via a tactical/strategic
+  memory split with conflict resolution before promotion.
+- Context engineering over prompt engineering: keep retrieved context
+  compact and just-in-time, give each tool a single clear job, and let tools
+  be polyglot processes behind a uniform JSON contract.
+
+## Quickstart
+
+```bash
+pip install fabri                    # the `fabri` console command lands on your PATH
+docker run -p 6333:6333 qdrant/qdrant   # vector store for the agent's memory
+export ANTHROPIC_API_KEY=...
+
+fabri init demo && cd demo           # scaffold a runnable starter project
+fabri --config agent.yaml run "greet Ada with the hello tool"
+```
+
+`fabri init` writes an `agent.yaml`, an example tool under `tools/agent_tools/`,
+and a `docker-compose.yml` — edit those, not the library. For OpenAI models,
+`pip install "fabri[openai]"` and set `llm.provider: openai`.
+
+## Setup (from a checkout)
+
+```bash
+docker compose up -d                 # starts Qdrant on :6333
+python3 -m venv .venv && .venv/bin/pip install -e ".[dev]"
+export ANTHROPIC_API_KEY=...         # only needed for AnthropicLLMBackend
+```
+
+Installing puts a `fabri` console command on your PATH — that's the entry point
+everything below uses. Embeddings run fully locally via
+`sentence-transformers/all-MiniLM-L6-v2` — no embedding API calls, no cloud
+dependency beyond the LLM itself.
+
+## Usage
+
+```bash
+fabri run "some task description"
+fabri --config agent.yaml run "some task description"   # config-driven agent, see docs/creating-an-agent.md
+fabri --verbose run "some task description"   # DEBUG logging to console too
+fabri inspect-memory "a query to test retrieval"
+fabri ingest-traces <session-id>
+```
+
+(`python cli.py ...` still works from a source checkout — it's a thin shim over
+the same `fabri.cli:main` the console script points at.)
+
+See `docs/creating-an-agent.md` for using this as a library/framework in
+another project: installing it, writing an `agent.yaml`, adding tools, and a
+worked example.
+
+`run` executes the agent loop, then immediately mines the resulting trace for
+failures and synthesizes any new guidelines. `ingest-traces` re-runs that
+synthesis step against a past session's trace on its own. `run` and
+`ingest-traces` both require `ANTHROPIC_API_KEY` (they fail fast with a clear
+message if it's unset, rather than a raw stack trace) — `inspect-memory` does
+not, since it only talks to Qdrant.
+
+Every run produces two records, both keyed by `session_id`: a structured
+JSONL trace (`.fabri/traces/<session_id>.jsonl`, the machine-readable
+record used by the pipeline) and a human-readable log
+(`.fabri/logs/<session_id>.log`, always
+DEBUG-level regardless of `--verbose` — that flag only controls what's also
+echoed to the console) with LLM call latency/token usage, tool dispatch
+latency, and every promotion/dedup decision the pipeline makes.
+
+Both land under `.fabri/` in the directory you run from (override with
+`$FABRI_HOME`), so each consuming project keeps its own traces and logs
+rather than scattering them into wherever the package happens to be installed.
+Add `.fabri/` to your project's `.gitignore`.
+
+Each run returns an `outcome`: `success` (clean), `success_with_recovery`
+(finished, but at least one tool call failed along the way), or `incomplete`
+(hit the step limit with no final answer).
+
+## Architecture
+
+```
+src/fabri/        # src/ layout: repo is "fabri", package is "fabri"
+  memory/        embeddings, schema, Qdrant store, compression, pruning/promotion
+  orchestrator/  retrieval, trace logging, the trace -> guideline pipeline
+  tools/         manifest schema, subprocess runner, example tools, agent-as-tool adapter
+  core/          the ReAct agent loop, decompose, and the pluggable LLM backend
+  runtime.py     build_llm/build_tools/build_tool_defs -- shared by cli and agent_runner_tool
+  admin.py       admin-only config inspection + dashboard (stub auth seam, see below)
+  config.py      YAML agent config loader (DEFAULT_CONFIG + load_config)
+  toon.py        TOON codec: compact JSON-shaped encoding for tool results (token savings)
+  paths.py       project-local .fabri/ resolution for traces + logs
+  cli.py         the `fabri` console command (composition over the public API)
+  __init__.py    public API: run_agent, load_config, ToolRegistry, QdrantMemoryStore, ...
+cli.py           thin shim: `python cli.py ...` -> fabri.cli:main
+```
+
+### Agents as tools
+
+A `tools.agents` entry in a config exposes a *different* agent.yaml as a tool
+on this one (`fabri/tools/agent_tool.py` builds the manifest,
+`agent_runner_tool.py` runs the sub-agent as an ordinary subprocess tool —
+stdin `{"task": ...}`, stdout `{"final_text", "outcome"}`). This is
+composition, not a new orchestrator: each sub-agent is just another tool call
+in the parent's normal ReAct loop. See `ludexel/.agent/game_content_agent.yaml`
+for a worked example wiring four domain agents (character/tiles/map/story)
+into one top-level agent.
+
+### Admin CLI / dashboard
+
+`cli.py admin config --config agent.yaml` prints the merged config plus the
+resolved tool registry as JSON; `cli.py admin dashboard --config agent.yaml`
+prints a human-readable summary (agent/llm/tools/memory counts). Both are
+gated by `require_admin()` (`fabri/admin.py`): if `FABRI_ADMIN_TOKEN`
+is unset the gate is open (no auth backend exists yet), but once it's set,
+`--admin-token` must match it. This is a placeholder seam for real auth
+(SSO/API gateway/etc.), not a security boundary by itself.
+
+### Tool contract
+
+A tool is a JSON manifest (`name`, `description`, `command`, schemas, timeout)
+next to an executable in any language. `runner.py` invokes `command` as a
+subprocess, writes the call's args as JSON to stdin, and parses stdout as
+JSON. The result is always normalized to `{ok, error?, result?, stderr?}`
+before the agent sees it — malformed output, timeouts, and tool error exits
+are distinct, well-defined failure modes rather than crashes. See
+`tools/examples/` for a Python tool (`echo`), a Go tool (`sum`), and a
+deliberately misbehaving tool (`broken`) used only by the test suite.
+
+### Token efficiency (TOON)
+
+Tool results are encoded into the model's context as **TOON**
+(Token-Oriented Object Notation, `toon.py`) rather than JSON by default — a
+compact indentation-based encoding that drops braces and collapses uniform
+arrays to a single header row plus data rows, typically ~30–40% fewer
+characters on tabular results. The framework encodes this itself (no model
+reliability risk); the trace/logs keep raw JSON. Pipeline: `json → toon → llm`
+inbound, and JSON outbound by default (`tools.result_format`,
+`agent.output_format` — see `docs/creating-an-agent.md`). Native tool-call
+arguments stay provider JSON. The codec round-trips any JSON-shaped value and
+`fabri.toon.encode`/`.decode` are public.
+
+### Memory lifecycle
+
+1. Every agent run logs a JSONL trace (`traces/<session_id>.jsonl`).
+2. `orchestrator/pipeline.py` scans a trace for tool-call failures and asks
+   the LLM to compress each into a short, generalized guideline
+   (`memory/compress.py` enforces a hard token cap regardless of what the
+   LLM returns).
+3. `memory/pruning.py` checks the new guideline against existing **tactical**
+   entries by cosine similarity. A near-duplicate increments that entry's
+   recurrence count instead of inserting a copy; once it has recurred across
+   3 distinct sessions (`PROMOTION_THRESHOLD_SESSIONS`), it's promoted to
+   **strategic**.
+4. `orchestrator/retrieval.py` embeds the next task, pulls the top-k most
+   relevant guidelines (tactical + strategic) from Qdrant, and formats them
+   as a compact bullet list injected into the agent's system prompt. If any
+   available tool's name appears in the task text, a second query filters by
+   that tool's tag (`MemoryEntry.tools`, populated in step 2 from whichever
+   tool actually failed) and those hits are *guaranteed* inclusion — this
+   surfaces tool-specific guidelines even when their wording is too
+   dissimilar from the query for cosine similarity alone to rank them in the
+   top-k. (A graph-DB-backed version of this was considered and rejected: the
+   only capability needed was a single-hop "guidelines for this tool" lookup,
+   which a payload filter gives you for free — a second database would have
+   added a dependency, a migration risk, and a way for the graph and the
+   vector store to drift out of sync, for no real capability gain.)
+
+This is what closes the loop: a failure in session N becomes retrievable
+context in session N+1, without a human re-writing the prompt by hand.
+
+## Known v1 limitations (intentional, not bugs)
+
+- **No embedding-model migration path.** Each memory point stores a
+  `model_version` payload field, but swapping the embedding model means
+  recreating the Qdrant collection from scratch — old vectors are not
+  re-embedded automatically.
+- **Single-writer assumption.** Concurrent *distinct* agent processes
+  ingesting guidelines at the same time are not locked against each other.
+  This is made safe-by-construction for the common case (same guideline
+  text) via deterministic point IDs (hash of the compressed text), but two
+  different writers racing on genuinely different new guidelines is
+  untested.
+- **No automatic eviction.** `memory/pruning.py:evict_stale` exists but is
+  not wired into a scheduled job — run it manually if the strategic store
+  grows large with low-value entries.
+
+## Tests
+
+```bash
+.venv/bin/python -m pytest tests/ -q
+```
+
+Covers store round-trip/idempotency, the dedup + promotion rule, the tool
+runner's normalized failure contract (Python + Go + malformed output +
+timeout), the token-cap enforcement in `compress.py`, the TOON codec
+(round-trip + fuzz), and the `fabri init` scaffold.
+
+## Releasing
+
+Tag a version (`git tag v0.1.0 && git push origin v0.1.0`) and GitHub Actions
+publishes to PyPI via Trusted Publishing — see `RELEASING.md` for the one-time
+setup.
+
+## License
+
+[Apache-2.0](LICENSE) © Rushikesh Patade.

fabri-0.1.0/README.md

@@ -0,0 +1,209 @@
+# fabri
+
+A local, open-source memory + orchestration layer for a custom LLM agent.
+Built around two ideas:
+
+- **SCOPE** (arXiv:2512.15374): treat the agent's prompt context as something
+  that evolves automatically from execution traces, via a tactical/strategic
+  memory split with conflict resolution before promotion.
+- Context engineering over prompt engineering: keep retrieved context
+  compact and just-in-time, give each tool a single clear job, and let tools
+  be polyglot processes behind a uniform JSON contract.
+
+## Quickstart
+
+```bash
+pip install fabri                    # the `fabri` console command lands on your PATH
+docker run -p 6333:6333 qdrant/qdrant   # vector store for the agent's memory
+export ANTHROPIC_API_KEY=...
+
+fabri init demo && cd demo           # scaffold a runnable starter project
+fabri --config agent.yaml run "greet Ada with the hello tool"
+```
+
+`fabri init` writes an `agent.yaml`, an example tool under `tools/agent_tools/`,
+and a `docker-compose.yml` — edit those, not the library. For OpenAI models,
+`pip install "fabri[openai]"` and set `llm.provider: openai`.
+
+## Setup (from a checkout)
+
+```bash
+docker compose up -d                 # starts Qdrant on :6333
+python3 -m venv .venv && .venv/bin/pip install -e ".[dev]"
+export ANTHROPIC_API_KEY=...         # only needed for AnthropicLLMBackend
+```
+
+Installing puts a `fabri` console command on your PATH — that's the entry point
+everything below uses. Embeddings run fully locally via
+`sentence-transformers/all-MiniLM-L6-v2` — no embedding API calls, no cloud
+dependency beyond the LLM itself.
+
+## Usage
+
+```bash
+fabri run "some task description"
+fabri --config agent.yaml run "some task description"   # config-driven agent, see docs/creating-an-agent.md
+fabri --verbose run "some task description"   # DEBUG logging to console too
+fabri inspect-memory "a query to test retrieval"
+fabri ingest-traces <session-id>
+```
+
+(`python cli.py ...` still works from a source checkout — it's a thin shim over
+the same `fabri.cli:main` the console script points at.)
+
+See `docs/creating-an-agent.md` for using this as a library/framework in
+another project: installing it, writing an `agent.yaml`, adding tools, and a
+worked example.
+
+`run` executes the agent loop, then immediately mines the resulting trace for
+failures and synthesizes any new guidelines. `ingest-traces` re-runs that
+synthesis step against a past session's trace on its own. `run` and
+`ingest-traces` both require `ANTHROPIC_API_KEY` (they fail fast with a clear
+message if it's unset, rather than a raw stack trace) — `inspect-memory` does
+not, since it only talks to Qdrant.
+
+Every run produces two records, both keyed by `session_id`: a structured
+JSONL trace (`.fabri/traces/<session_id>.jsonl`, the machine-readable
+record used by the pipeline) and a human-readable log
+(`.fabri/logs/<session_id>.log`, always
+DEBUG-level regardless of `--verbose` — that flag only controls what's also
+echoed to the console) with LLM call latency/token usage, tool dispatch
+latency, and every promotion/dedup decision the pipeline makes.
+
+Both land under `.fabri/` in the directory you run from (override with
+`$FABRI_HOME`), so each consuming project keeps its own traces and logs
+rather than scattering them into wherever the package happens to be installed.
+Add `.fabri/` to your project's `.gitignore`.
+
+Each run returns an `outcome`: `success` (clean), `success_with_recovery`
+(finished, but at least one tool call failed along the way), or `incomplete`
+(hit the step limit with no final answer).
+
+## Architecture
+
+```
+src/fabri/        # src/ layout: repo is "fabri", package is "fabri"
+  memory/        embeddings, schema, Qdrant store, compression, pruning/promotion
+  orchestrator/  retrieval, trace logging, the trace -> guideline pipeline
+  tools/         manifest schema, subprocess runner, example tools, agent-as-tool adapter
+  core/          the ReAct agent loop, decompose, and the pluggable LLM backend
+  runtime.py     build_llm/build_tools/build_tool_defs -- shared by cli and agent_runner_tool
+  admin.py       admin-only config inspection + dashboard (stub auth seam, see below)
+  config.py      YAML agent config loader (DEFAULT_CONFIG + load_config)
+  toon.py        TOON codec: compact JSON-shaped encoding for tool results (token savings)
+  paths.py       project-local .fabri/ resolution for traces + logs
+  cli.py         the `fabri` console command (composition over the public API)
+  __init__.py    public API: run_agent, load_config, ToolRegistry, QdrantMemoryStore, ...
+cli.py           thin shim: `python cli.py ...` -> fabri.cli:main
+```
+
+### Agents as tools
+
+A `tools.agents` entry in a config exposes a *different* agent.yaml as a tool
+on this one (`fabri/tools/agent_tool.py` builds the manifest,
+`agent_runner_tool.py` runs the sub-agent as an ordinary subprocess tool —
+stdin `{"task": ...}`, stdout `{"final_text", "outcome"}`). This is
+composition, not a new orchestrator: each sub-agent is just another tool call
+in the parent's normal ReAct loop. See `ludexel/.agent/game_content_agent.yaml`
+for a worked example wiring four domain agents (character/tiles/map/story)
+into one top-level agent.
+
+### Admin CLI / dashboard
+
+`cli.py admin config --config agent.yaml` prints the merged config plus the
+resolved tool registry as JSON; `cli.py admin dashboard --config agent.yaml`
+prints a human-readable summary (agent/llm/tools/memory counts). Both are
+gated by `require_admin()` (`fabri/admin.py`): if `FABRI_ADMIN_TOKEN`
+is unset the gate is open (no auth backend exists yet), but once it's set,
+`--admin-token` must match it. This is a placeholder seam for real auth
+(SSO/API gateway/etc.), not a security boundary by itself.
+
+### Tool contract
+
+A tool is a JSON manifest (`name`, `description`, `command`, schemas, timeout)
+next to an executable in any language. `runner.py` invokes `command` as a
+subprocess, writes the call's args as JSON to stdin, and parses stdout as
+JSON. The result is always normalized to `{ok, error?, result?, stderr?}`
+before the agent sees it — malformed output, timeouts, and tool error exits
+are distinct, well-defined failure modes rather than crashes. See
+`tools/examples/` for a Python tool (`echo`), a Go tool (`sum`), and a
+deliberately misbehaving tool (`broken`) used only by the test suite.
+
+### Token efficiency (TOON)
+
+Tool results are encoded into the model's context as **TOON**
+(Token-Oriented Object Notation, `toon.py`) rather than JSON by default — a
+compact indentation-based encoding that drops braces and collapses uniform
+arrays to a single header row plus data rows, typically ~30–40% fewer
+characters on tabular results. The framework encodes this itself (no model
+reliability risk); the trace/logs keep raw JSON. Pipeline: `json → toon → llm`
+inbound, and JSON outbound by default (`tools.result_format`,
+`agent.output_format` — see `docs/creating-an-agent.md`). Native tool-call
+arguments stay provider JSON. The codec round-trips any JSON-shaped value and
+`fabri.toon.encode`/`.decode` are public.
+
+### Memory lifecycle
+
+1. Every agent run logs a JSONL trace (`traces/<session_id>.jsonl`).
+2. `orchestrator/pipeline.py` scans a trace for tool-call failures and asks
+   the LLM to compress each into a short, generalized guideline
+   (`memory/compress.py` enforces a hard token cap regardless of what the
+   LLM returns).
+3. `memory/pruning.py` checks the new guideline against existing **tactical**
+   entries by cosine similarity. A near-duplicate increments that entry's
+   recurrence count instead of inserting a copy; once it has recurred across
+   3 distinct sessions (`PROMOTION_THRESHOLD_SESSIONS`), it's promoted to
+   **strategic**.
+4. `orchestrator/retrieval.py` embeds the next task, pulls the top-k most
+   relevant guidelines (tactical + strategic) from Qdrant, and formats them
+   as a compact bullet list injected into the agent's system prompt. If any
+   available tool's name appears in the task text, a second query filters by
+   that tool's tag (`MemoryEntry.tools`, populated in step 2 from whichever
+   tool actually failed) and those hits are *guaranteed* inclusion — this
+   surfaces tool-specific guidelines even when their wording is too
+   dissimilar from the query for cosine similarity alone to rank them in the
+   top-k. (A graph-DB-backed version of this was considered and rejected: the
+   only capability needed was a single-hop "guidelines for this tool" lookup,
+   which a payload filter gives you for free — a second database would have
+   added a dependency, a migration risk, and a way for the graph and the
+   vector store to drift out of sync, for no real capability gain.)
+
+This is what closes the loop: a failure in session N becomes retrievable
+context in session N+1, without a human re-writing the prompt by hand.
+
+## Known v1 limitations (intentional, not bugs)
+
+- **No embedding-model migration path.** Each memory point stores a
+  `model_version` payload field, but swapping the embedding model means
+  recreating the Qdrant collection from scratch — old vectors are not
+  re-embedded automatically.
+- **Single-writer assumption.** Concurrent *distinct* agent processes
+  ingesting guidelines at the same time are not locked against each other.
+  This is made safe-by-construction for the common case (same guideline
+  text) via deterministic point IDs (hash of the compressed text), but two
+  different writers racing on genuinely different new guidelines is
+  untested.
+- **No automatic eviction.** `memory/pruning.py:evict_stale` exists but is
+  not wired into a scheduled job — run it manually if the strategic store
+  grows large with low-value entries.
+
+## Tests
+
+```bash
+.venv/bin/python -m pytest tests/ -q
+```
+
+Covers store round-trip/idempotency, the dedup + promotion rule, the tool
+runner's normalized failure contract (Python + Go + malformed output +
+timeout), the token-cap enforcement in `compress.py`, the TOON codec
+(round-trip + fuzz), and the `fabri init` scaffold.
+
+## Releasing
+
+Tag a version (`git tag v0.1.0 && git push origin v0.1.0`) and GitHub Actions
+publishes to PyPI via Trusted Publishing — see `RELEASING.md` for the one-time
+setup.
+
+## License
+
+[Apache-2.0](LICENSE) © Rushikesh Patade.